User intention analysis system and method based on artificial intelligence

Abstract

The invention relates to the technical field of user intention analysis, in particular to a user intention analysis system and method based on artificial intelligence, and the method comprises the steps: generating a single-point intention recognition result representing the current instantaneous operation of a user through multi-modal behavior data at the current moment; generating a time sequence intention chain representing continuous operation of the user according to the multi-modal behavior data at the historical moment; when an intention conflict is detected, calculating an optimal transfer path for transferring a current conflict intention node in the time sequence intention chain to a candidate target intention node according to a transfer probability weight in a pre-constructed intention transfer probability graph; and determining a resource demand corresponding to the candidate target intention node of the optimal transfer path, dynamically allocating resources to the candidate target intention node according to the current available resource state and the resource demand, and performing intention analysis. Resource allocation is dynamically optimized after intention conflicts are effectively detected and solved, and accurate understanding and efficient execution of user intentions are achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of user intention analysis, in particular to a user intention analysis system and method based on artificial intelligence. BACKGROUND

[0002] In the current booming development of artificial intelligence technology, user intention analysis has become the core function of many intelligent interaction systems. This function aims to accurately understand the current needs and intentions of users through deep mining and analysis of user behavior data, and to provide personalized and intelligent service responses for users.

[0003] However, in practical applications, the current user intention analysis technology faces many challenges. User behavior is highly dynamic and complex, so user intentions will produce intention conflicts. The user intention analysis process requires a large amount of computing resources, such as processor power and memory. Therefore, when intention conflicts occur, how to reasonably and dynamically allocate computing resources becomes a key problem. Existing technologies mostly use static or simple allocation strategies in resource allocation, which cannot accurately adjust according to the actual needs of the current intention analysis task and the real-time state of the resources, resulting in low resource utilization and seriously affecting the overall operation efficiency and performance of the system.

[0004] To solve the above problems, the present application provides a user intention analysis system and method based on artificial intelligence, which can realize more accurate, efficient and reasonable resource utilization of user intention analysis, and significantly improve the performance and user experience of intelligent interaction systems. SUMMARY

[0005] (1) Technical problem to be solved

[0006] The purpose of the present application is to provide a user intention analysis system and method based on artificial intelligence to solve the problem of not being able to effectively detect and solve intention conflicts and dynamically optimize resource allocation after the conflicts, and to realize accurate understanding and efficient execution of user intentions in complex interaction scenarios.

[0007] (2) Technical solution

[0008] To achieve the above purpose, on the one hand, the present application provides a user intention analysis system based on artificial intelligence, which comprises: a multi-modal data acquisition and intention generation module, an intention conflict detection and optimal transfer path calculation module, a resource allocation and intention analysis execution module, and the modules are sequentially connected in communication:

[0009] A multi-modal data acquisition and intention generation module is configured to acquire multi-modal behavior data generated by a user at a current time and historical times, the multi-modal behavior data including voice data, image data, and text data; generate a single-point intention recognition result representing a current instantaneous operation of the user based on the multi-modal behavior data at the current time; and generate a time-series intention chain representing a continuous operation of the user based on the multi-modal behavior data at the historical times.

[0010] An intention conflict detection and optimal transition path calculation module is configured to detect whether an intention conflict exists at the current time, the intention conflict being a conflict between the single-point intention recognition result and a current active intention node in the time-series intention chain; and calculate an optimal transition path from a current conflict intention node in the time-series intention chain to a candidate target intention node based on transition probability weights in a pre-constructed intention transition probability graph when the intention conflict is detected.

[0011] A resource allocation and intention analysis execution module is configured to determine resource requirements corresponding to the candidate target intention node of the optimal transition path, and to dynamically allocate resources to the candidate target intention node and perform intention analysis based on a current available resource state and the resource requirements.

[0012] Further, the intention conflict detection and optimal transition path calculation module includes the following steps:

[0013] A semantic feature vector of the single-point intention recognition result and a semantic feature vector of the current active intention node are extracted, and a semantic similarity value between the semantic feature vectors is calculated.

[0014] A duration weight coefficient of the current active intention node in the time-series intention chain is acquired; when the semantic similarity value is less than a preset conflict determination threshold, it is determined that an intention conflict exists; when it is determined that an intention conflict exists, a conflict confidence value is calculated based on the semantic similarity value and the duration weight coefficient; and if the conflict confidence value exceeds a preset conflict confidence threshold, the intention conflict is confirmed.

[0015] Further, the intention conflict detection and optimal transition path calculation module includes:

[0016] A candidate target intention node set determination module is configured to acquire a node set in a pre-constructed intention transition probability graph, directed edges connecting the nodes, and transition probability weights corresponding to the directed edges; and determine a candidate target intention node set based on a pre-defined screening rule with the current conflict intention node in the time-series intention chain as a path starting point.

[0017] a maximum cumulative probability and predecessor node analysis module, configured to calculate, according to the transition probability weight value, a maximum cumulative probability value and a corresponding predecessor node from the path starting point to each node in the candidate target intent node set by using a dynamic programming algorithm.

[0018] an optimal transition path calculation module, configured to traverse the candidate target intent node set, select a node with the maximum cumulative probability value as an optimal target intent node, and perform reverse backtracking according to the predecessor node corresponding to the optimal target intent node to obtain a complete transition path from the path starting point to the optimal target intent node as an optimal transition path.

[0019] Further, the maximum cumulative probability and predecessor node analysis module includes the following steps:

[0020] The maximum cumulative probability value corresponding to the path starting point is initialized as a reference value, and the path starting point is marked as having no predecessor node; the maximum cumulative probability values of other nodes are initialized as invalid values; and all nodes are traversed in a topological order of a pre-constructed intent transition probability graph.

[0021] For a currently traversed node, all predecessor nodes corresponding to all incoming edges of the currently traversed node are traversed, and a product of the maximum cumulative probability value of the predecessor node and the transition probability weight value of the corresponding incoming edge is calculated as a candidate cumulative probability value; the incoming edge is a directed edge in the intent transition probability graph pointing to the currently traversed node.

[0022] The maximum value of the candidate cumulative probability value is extracted as the maximum cumulative probability value of the currently traversed node, and the predecessor node corresponding to the incoming edge generating the maximum cumulative probability value is recorded as the predecessor node of the currently traversed node.

[0023] The above traversal process is repeated to obtain the maximum cumulative probability value and the corresponding predecessor node of each node in the candidate target intent node set.

[0024] Further, the resource allocation and intent analysis execution module includes:

[0025] a basic resource demand vector calculation module, configured to retrieve an intent resource mapping table according to a type identifier of a candidate target intent node to obtain a basic resource demand vector;

[0026] a resource allocation module, configured to scan real-time load states of each computing unit in a resource pool to obtain a multi-dimensional available resource vector, obtain a resource allocation scheme by using a resource scheduling optimization method on the basic resource demand vector and the multi-dimensional available resource vector, update the resource pool after preemptively reclaiming resources occupied by non-active intent nodes according to the resource allocation scheme, allocate resources to the candidate target intent node according to the updated resource pool, and perform intent analysis on the analysis task after the resource allocation is completed.

[0027] Further, the basic resource demand vector calculation module comprises the following steps:

[0028] Access the intent resource mapping table with the type identifier of the candidate target intent node as the query key, the intent resource mapping table storing in the form of key-value pairs with the type identifier of the intent node as the key and the corresponding basic resource demand description as the value.

[0029] Find the entry that exactly matches the query key in the intent resource mapping table; extract the stored basic resource demand description from the matched entry; and obtain the basic resource demand vector after parsing the extracted basic resource demand description.

[0030] Further, the resource allocation module comprises:

[0031] A resource allocation optimization model construction module is configured to establish a resource allocation optimization model with the goal of maximizing the satisfaction of the resource demand of the candidate target intent node and maximizing the resource utilization rate; the basic resource demand vector is used as the demand input constraint of the resource allocation optimization model, and the multi-dimensional available resource vector is used as the resource supply constraint of the resource allocation optimization model.

[0032] A resource allocation scheme generation module is configured to solve the resource allocation optimization model to obtain a candidate allocation scheme set; calculate the resource utilization efficiency evaluation value corresponding to each candidate allocation scheme in the candidate allocation scheme set; and select the candidate allocation scheme with the highest resource utilization efficiency evaluation value as the final resource allocation scheme.

[0033] Further, the resource allocation scheme generation module comprises the following steps:

[0034] Obtain the actual allocation resource vector by obtaining the resource type and the number of allocated resources for the candidate target intent node in the candidate allocation scheme; and obtain the basic resource demand vector corresponding to the candidate target intent node and the current multi-dimensional available resource vector of the resource pool.

[0035] Calculate the resource demand matching degree according to the actual allocation resource vector and the basic resource demand vector; and calculate the resource utilization rate according to the actual allocation resource vector and the multi-dimensional available resource vector.

[0036] Calculate the resource utilization efficiency evaluation value by weighting the resource demand matching degree and the resource utilization rate.

[0037] Further, the resource allocation module further comprises the following steps:

[0038] When the resource allocation scheme determines that the current multi-dimensional available resource vector cannot meet the basic resource demand vector of the candidate target intent node, all occupied resource intent nodes in the resource pool are scanned, the operation timestamp sequence of the intent node within a predefined time window is extracted, the variance of adjacent operation time intervals is calculated as the behavior dispersion, and the difference between the current time and the last valid operation timestamp of the intent node is obtained as the silence duration.

[0039] The behavior dispersion and the silence duration are obtained through a preset inactivity determination rule to obtain an activity score value, when the activity score value is lower than a preset activity score threshold, the intent node is marked as an inactive intent node to be recycled, the inactive intent nodes are sorted in descending order according to the total amount of occupied resources, and the resources of the inactive intent nodes are recycled into the resource pool in sequence.

[0040] On the other hand, based on the same inventive concept, the present application also provides an artificial intelligence-based user intent analysis method, which comprises;

[0041] Obtaining multi-modal behavior data generated by the user at the current time and the historical time, the multi-modal behavior data including voice data, image data and text data; generating a single-point intent recognition result representing the user's current instantaneous operation according to the multi-modal behavior data at the current time; and generating a time-series intent chain representing the user's continuous operation according to the multi-modal behavior data at the historical time.

[0042] Detecting whether there is an intent conflict, the intent conflict being a conflict between the single-point intent recognition result and a currently active intent node in the time-series intent chain; when an intent conflict is detected, calculating an optimal transfer path from the currently conflicting intent node in the time-series intent chain to a candidate target intent node according to the transfer probability weight in the pre-constructed intent transfer probability graph.

[0043] Determining the resource demand corresponding to the candidate target intent node of the optimal transfer path, and dynamically allocating resources to the candidate target intent node according to the current available resource state and the resource demand and performing intent analysis.

[0044] (3) Advantageous effects

[0045] Compared with the prior art, the present application has the following advantages:

[0046] 1. By acquiring the current time and historical time multi-modal behavior data of the user, single-point intent recognition results and time-series intent chains are generated respectively. By using the method of extracting semantic feature vectors, calculating semantic similarity values, and combining the duration weight coefficient to calculate the conflict confidence value, the conflict between the single-point intent recognition results and the current active intent node in the time-series intent chain is accurately detected, which can sharply capture the dynamic changes of user intent, avoid analysis deviation caused by ignoring intent conflict, and significantly improve the intent recognition accuracy in complex scenarios.

[0047] 2. After detecting the intent conflict, the dynamic programming algorithm is used to calculate the optimal transition path from the current conflict intent node to the candidate target intent node according to the pre-constructed intent transition probability graph. Not only does it ensure that the user's intent can smoothly and continuously transition when there is a conflict, avoiding intent interruption or confusion, but it also improves the system's efficiency in handling complex intent changes, allowing the system to quickly respond and adjust to the appropriate intent state, providing a smoother interactive experience for the user.

[0048] 3. After determining the optimal transition path, the resource allocation scheme is generated by retrieving the basic resource requirement vector from the intent resource mapping table according to the type identifier of the candidate target intent node, and scanning the resource pool to obtain the multi-dimensional available resource vector. By analyzing the behavior dispersion and silence duration of the intent node, non-active intent nodes are determined and recycled. According to the actual demand, reasonable allocation of resources is avoided, and the resource utilization rate is improved, while ensuring the stable operation of the system in different intent states, improving the overall system performance. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 The module composition diagram of the user intent analysis system based on artificial intelligence of embodiment 2 of the present application.

[0050] Figure 2 The flowchart of the user intent analysis method based on artificial intelligence of embodiment 1 of the present application. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be described in detail below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor are within the scope of protection of the present application.

[0052] Before providing examples, it is necessary to describe the application scenarios of the present invention. The present invention is a user intent analysis system and method based on artificial intelligence, which is applied in complex interaction scenarios to dynamically optimize resource allocation after analyzing user intent conflicts, so as to achieve accurate understanding and efficient execution of user intent.

[0053] Example 1: As Figure 1 As shown, this embodiment provides an artificial intelligence-based user intent analysis system. The system includes: a multimodal data acquisition and intent generation module, an intent conflict detection and optimal transfer path calculation module, and a resource allocation and intent analysis execution module. These modules are sequentially connected in communication.

[0054] A multi-modal data acquisition and intention generation module is configured to acquire multi-modal behavior data generated by a user at a current time and historical times, the multi-modal behavior data including voice data, image data, and text data; generate a single-point intention recognition result representing a current instantaneous operation of the user according to the multi-modal behavior data at the current time; and generate a time-series intention chain representing a continuous operation of the user according to the multi-modal behavior data at the historical times. For example, the user intention analysis of the embodiment is a smart home control system. When the user interacts with the system through various means such as voice assistants, mobile applications, touch panels, etc., the system needs to accurately understand the user's true intention and provide corresponding service responses. The system first establishes a comprehensive multi-modal data acquisition mechanism. Voice data acquisition is achieved through microphone arrays distributed throughout the room, which can capture the user's voice commands in real time, such as "turn up the lights in the living room a bit" or "play some light music". Image data acquisition is completed through camera devices installed at key locations. These cameras can not only recognize the user's facial expressions and body movements, but also analyze the user's location movement trajectory and dwell time. Text data acquisition is mainly obtained through the user's input operations on the mobile application or wall touch panel, including text input, button clicks, slide selection, and other interactive behaviors. The system labels each collected data with an accurate timestamp, data type, acquisition device identifier, and preliminary confidence score to ensure accurate data source tracing in subsequent processing. For the multi-modal behavior data collected at the current time, the system runs a single-point intention recognition algorithm to generate an instantaneous intention recognition result, i.e., a single-point intention recognition result. For voice data, the system first converts audio signals into text content through automatic speech recognition technology, then uses natural language processing algorithms to extract semantic features, and identifies keywords, syntax structures, and emotional colors. For image data, the system uses computer vision algorithms to analyze user action patterns, such as the user walking into a room indicating the use of devices in that room, and user hand gestures indicating the need to adjust environmental parameters. For text data, the system directly performs semantic analysis and intention classification. These different modal feature information is comprehensively processed through a multi-modal fusion algorithm, which assigns different weights according to the reliability and relevance of each modal data, and finally generates a unified single-point intention recognition result, such as identifying the specific intention category of "environmental light adjustment" with corresponding parameter information such as target brightness value. In parallel with single-point intention recognition is the construction and maintenance of the time-series intention chain. The system continuously tracks the user's behavior data at historical times, and constructs an intention sequence reflecting the user's continuous operation mode according to the time sequence. Each node in the time-series intention chain represents a user's intention operation at a specific time period, and the node contains detailed information such as intention type, start time, end time, related parameters, and execution status.For example, the sequence of user's behavior after coming home in the evening can include "turn on the porch light", "adjust the temperature in the living room", "start the sound system", "check the security monitoring", and so on. The system will determine whether these intents constitute a coherent behavior chain according to the time interval between operations and semantic relevance, and identify which intents are currently still in the active execution state. The construction of the time sequence intent chain adopts a sliding time window mechanism, usually retaining user behavior data within the past two hours, and archiving historical data beyond the time window, which does not participate in the current intent analysis process.

[0055] An intent conflict detection and optimal transfer path calculation module is configured to detect whether there is an intent conflict between the single-point intent recognition result and the current active intent node in the time sequence intent chain; when an intent conflict is detected, the optimal transfer path from the current conflict intent node in the time sequence intent chain to the candidate target intent node is calculated according to the transfer probability weight in the pre-constructed intent transfer probability graph;

[0056] A resource allocation and intent analysis execution module is configured to determine the resource demand corresponding to the candidate target intent node of the optimal transfer path, and dynamically allocate resources to the candidate target intent node and perform intent analysis according to the current available resource state and the resource demand.

[0057] The intent conflict detection and optimal transfer path calculation module includes the following steps:

[0058] extract the semantic feature vectors of the single-point intent recognition result and the current active intent node, and calculate the semantic similarity value between the semantic feature vectors; in the intent conflict detection process of the smart home control system, the system needs to determine whether there is a real intent conflict situation through accurate numerical calculation and logical judgment. When the system obtains the single-point intent recognition result "adjust the brightness of the living room lights" and the current active intent node "watching movie mode", first, the two intents need to be deeply semantic feature extracted and similarity analyzed. The extraction process of the semantic feature vector adopts a pre-trained deep learning semantic encoding model, which can convert the intent described in natural language into a 512-dimensional dense vector representation. For the single-point intent "adjust the brightness of the living room lights", the system first performs text preprocessing, including word segmentation, part-of-speech tagging, named entity recognition and other steps, and then inputs the processed text into the semantic encoder to obtain the corresponding semantic feature vector. Similarly, the system also performs the same feature extraction process on the current active intent node "watching movie mode" to generate the corresponding semantic feature vector. The two semantic feature vectors can accurately reflect the semantic content and context association information of their respective intents in the high-dimensional semantic space. The system then calculates the cosine similarity value between the two semantic feature vectors, and the calculation process is realized by dividing the product of the vector dot product by the product of the vector module. Assuming that the calculated semantic similarity value is 0.72, the system compares this value with the preset conflict determination threshold value 0.75. The setting of the conflict determination threshold value is based on the statistical analysis of a large amount of historical data and the optimization results of the system performance, which not only ensures that the real intent conflict can be effectively identified, but also avoids excessive sensitivity leading to misjudgment.

[0059] obtaining a duration weight coefficient of the current active intent node in the time sequence intent chain; determining that there is an intent conflict when the semantic similarity value is less than a preset conflict determination threshold; calculating a conflict confidence value according to the semantic similarity value and the duration weight coefficient when it is determined that there is an intent conflict; and confirming the intent conflict when the conflict confidence value exceeds a preset conflict confidence threshold. When the similarity value 0.72 is less than the preset conflict determination threshold 0.75, the system preliminarily determines that there is a potential intent conflict. In order to further confirm the reliability of the conflict, the system needs to obtain the duration weight coefficient of the current active intent node in the time sequence intent chain. The calculation of the duration weight coefficient uses an exponential decay function, which takes into account the time stability characteristics of the intent, that is, the longer an intent lasts, the more important it is in the user behavior pattern, and the corresponding weight coefficient is also larger. For example, the "watching movie mode" intent has lasted for 25 minutes, and according to the system preset weight calculation formula, its corresponding duration weight coefficient is 0.85, reflecting the stability and importance of the intent in the current user behavior sequence. The system then calculates the conflict confidence value according to the semantic similarity value and the duration weight coefficient. The calculation of the conflict confidence value uses a weighted product method, that is, the semantic similarity value 0.72 is multiplied by the duration weight coefficient 0.85 to obtain the conflict confidence value 0.61. The conflict confidence value reflects the severity and reliability of the intent conflict, taking into account both the semantic level of the conflict and the stability factor in the time dimension. Since the conflict confidence value 0.61 is greater than the preset conflict confidence threshold 0.5, the system finally confirms that there is indeed an intent conflict that needs to be handled.

[0060] The intent conflict detection and optimal transfer path calculation module comprises:

[0061] The candidate target intent node set determination module is configured to acquire a node set, a directed edge connecting the nodes, and a transition probability weight corresponding to each directed edge in a pre-constructed intent transition probability graph; and determine a candidate target intent node set according to a pre-defined screening rule, with a current conflict intent node in the time sequence intent chain as a path starting point. The pre-constructed intent transition probability graph is a core knowledge base of the system, and is constructed by analyzing historical behavior data of tens of thousands of users, and contains all common intent types and their transition relationships in the smart home environment. The node set in the graph covers hundreds of specific intent types such as "light control", "temperature regulation", "entertainment playing", "security monitoring", and "environmental perception", each node has a unique identifier and detailed attribute description. The directed edge connecting the nodes represents the transition relationship that may occur between intents, the direction of the edge represents the directionality of the transition, and the weight value marked on the edge represents the probability of the transition, the weight value ranges from 0 to 1, and the larger the value is, the higher the transition probability is. The system first takes the current conflict intent node "movie watching mode" in the time sequence intent chain as the starting point of path search, and then determines a candidate target intent node set according to a pre-defined screening rule. The screening rule includes multiple dimensions of judgment criteria, first, semantic correlation screening, that is, the candidate target intent must have a certain relevance in semantics with the current detected single-point intent "adjusting the brightness of the living room light". Secondly, user preference screening, the system will exclude intent options that do not conform to the user's habits according to the user's historical behavior patterns and personal preference settings. Finally, context rationality screening, that is, the candidate intent must conform to the current environmental state and device availability. After screening, the system determines three candidate target intent nodes, which are "movie watching light mode", "pause movie to adjust light", and "end movie to restore normal lighting".

[0062] The maximum cumulative probability and predecessor node analysis module is configured to calculate, according to the transition probability weight value, a maximum cumulative probability value and a corresponding predecessor node from the path starting point to each node in the candidate target intent node set by using a dynamic programming algorithm.

[0063] An optimal transition path calculation module is configured to traverse the candidate target intent node set, select a node with the largest cumulative probability value as an optimal target intent node, and perform backtracking on a predecessor node corresponding to the optimal target intent node to obtain a complete transition path from the path starting point to the optimal target intent node as an optimal transition path. During the calculation process, the system finds that the maximum cumulative probability value of reaching the "movie watching light mode" node is 0.72, and the optimal path is "movie watching mode -> adjusting environmental parameters -> movie watching light mode". The maximum cumulative probability value of reaching the "pause movie light adjustment" node is 0.65, and the optimal path is "movie watching mode -> pause movie light adjustment". The maximum cumulative probability value of reaching the "end movie normal lighting" node is 0.48, and the optimal path is "movie watching mode -> user activity switching -> end movie normal lighting". The system traverses all candidate target intent nodes, compares their maximum cumulative probability values, and selects the "movie watching light mode" node with the highest probability as the optimal target intent node. After determining the optimal target intent node, the system performs backtracking according to the predecessor node information recorded during the dynamic programming process. Starting from the "movie watching light mode" node, the system finds its predecessor node "adjusting environmental parameters", then continues to find the predecessor node "movie watching mode" of "adjusting environmental parameters", and finally backtracks to the starting node. Through this backtracking process, the system reconstructs the complete optimal transition path "movie watching mode -> adjusting environmental parameters -> movie watching light mode", which represents the most likely direction of user intent evolution.

[0064] The maximum cumulative probability and predecessor node analysis module includes the following steps:

[0065] initializing the maximum cumulative probability value corresponding to the path start node as a reference value, and marking it as having no predecessor node; initializing the maximum cumulative probability value of other nodes as an invalid value; traversing all nodes in topological order of the pre-constructed intention transition probability graph; the system first initializes all nodes in the intention transition probability graph. For the path start node "movie watching mode", the system sets its maximum cumulative probability value to the reference value 1.0, indicating that the probability from the node to itself is a certain value, and at the same time marks that the node has no predecessor node, and marks the predecessor node record as null. For all other nodes in the intention transition probability graph, the system initializes their maximum cumulative probability values to negative infinity, indicating that an effective path has not yet been found, and the predecessor node information is also initialized to an undefined state. The system then traverses all nodes in topological order of the pre-constructed intention transition probability graph. Topological sorting is a linear ordering of directed acyclic graphs, which ensures that for any directed edge in the intention transition probability graph, the starting node appears before the ending node in the ordering. This ordering ensures that when a node is processed in the dynamic programming process, all predecessor nodes that may affect the optimal solution of the node have been correctly processed, thereby ensuring the correctness and efficiency of the algorithm. The traversal process processes each node in the order of the topological sorting result.

[0066] For the current traversal node, traverse all predecessor nodes corresponding to the current traversal node all incoming edges, and calculate the product of the maximum cumulative probability value of the predecessor node and the transition probability weight value of the corresponding incoming edge as the candidate cumulative probability value; the incoming edge is a directed edge in the intention transition probability graph pointing to the current traversal node; when the system traverses a certain node, such as the "adjust environmental parameters" node, it first needs to identify all incoming edges and corresponding predecessor nodes of the node. The incoming edge is a directed edge in the intention transition probability graph pointing to the current traversal node, each incoming edge connects a predecessor node and the current node, and is marked with a corresponding transition probability weight value. For the "adjust environmental parameters" node, the system finds two incoming edges, one from the "movie watching mode" node (transition probability weight 0.8) and the other from the "music playing mode" node (transition probability weight 0.6). The system then calculates the candidate cumulative probability value corresponding to each incoming edge. The calculation method of the candidate cumulative probability value is to multiply the current maximum cumulative probability value of the predecessor node and the transition probability weight value of the corresponding incoming edge. For the incoming edge from the "movie watching mode", the candidate cumulative probability value is 1.0 multiplied by 0.8, equal to 0.8. For the incoming edge from the "music playing mode", since the maximum cumulative probability value of the "music playing mode" node is still the initial negative infinity value, indicating that an effective path has not yet been found, the corresponding candidate cumulative probability value is also negative infinity, which is excluded in actual calculation.

[0067] The maximum value of the candidate accumulated probability value is extracted as the maximum accumulated probability value of the current traversal node, and the predecessor node corresponding to the incoming edge generating the maximum accumulated probability value is recorded as the predecessor node of the current traversal node; the system extracts the maximum value from all valid candidate accumulated probability values as the maximum accumulated probability value of the current traversal node. In this example, the "adjust environmental parameters" node has only one valid candidate value 0.8, so this value is set as the maximum accumulated probability value of the node. At the same time, the system records the predecessor node "movie watching mode" corresponding to the incoming edge generating the maximum accumulated probability value as the predecessor node of the "adjust environmental parameters" node, and this predecessor node information will play an important role in the subsequent path backtracking process.

[0068] The above traversal process is repeated to obtain the maximum accumulated probability value and the corresponding predecessor node of each node in the candidate target intent node set. The traversal and calculation process continues until all nodes in the topological sorting are processed. Through this systematic processing method, each node can obtain the maximum accumulated probability value corresponding to the optimal path from the starting node to the node, and the direct predecessor node information of the node in the optimal path. Finally, the system obtains the complete optimal path information of each node in the candidate target intent node set.

[0069] The resource allocation and intent analysis execution module includes:

[0070] The basic resource demand vector calculation module is used to retrieve the basic resource demand vector from the intent resource mapping table according to the type identifier of the candidate target intent node;

[0071] The resource allocation module is used to scan the real-time load state of each computing unit in the resource pool to obtain a multi-dimensional available resource vector; the basic resource demand vector and the multi-dimensional available resource vector are combined through a resource scheduling optimization method to obtain a resource allocation scheme; the resources occupied by the non-active intent nodes are preemptively recovered according to the resource allocation scheme, and the resource pool is updated; the candidate target intent nodes are allocated resources according to the updated resource pool; and the intent analysis is performed on the analysis task after the resource allocation is completed. After the resource allocation is completed, the system starts the intent analysis processing flow according to the determined resource allocation scheme. The intent analysis processing flow initializes the computing environment allocated to the task, including creating an independent process space, allocating a specified memory area, establishing a network communication channel, and preparing necessary data storage space, etc. Then the system loads the corresponding intent analysis algorithm module, calls the corresponding processing logic according to the specific intent type "movie watching light mode", performs specific operations such as environment perception, device control, parameter adjustment, etc., and finally realizes accurate identification and intelligent response of user intent.

[0072] The basic resource demand vector calculation module includes the following steps:

[0073] The type identifier of the candidate target intent node is accessed as a query key to an intent resource mapping table, which stores key-value pairs with type identifiers of intent nodes as keys and corresponding basic resource requirement descriptions as values. In the resource requirement determination process, the system needs to obtain the basic resource configuration information required by the candidate target intent node through accurate data query and parsing operations. After the system determines that the type identifier of the candidate target intent node is "MOVIE_LIGHTING_MODE", the identifier is immediately used as a query key value to start the access process of the intent resource mapping table. The intent resource mapping table is the core data structure of the system resource management subsystem, which organizes data in a key-value pair storage manner and has the characteristics of fast query and efficient access. The keys in the intent resource mapping table are type identifiers of various intent nodes, which adopt a unified naming specification and can uniquely determine each intent type. The values in the intent resource mapping table are detailed resource requirement descriptions corresponding to the intent types, which contain all the computing resource information required to execute the analysis of the intent type. The data of the intent resource mapping table comes from performance testing in the system design phase and statistical analysis during operation, ensuring the accuracy and practicality of the resource requirement description.

[0074] In the intent resource mapping table, an entry that exactly matches the query key is searched for; the stored base resource requirement description is extracted from the matched entry; and the extracted base resource requirement description is parsed to obtain a base resource requirement vector. The system performs an exact match search operation in the intent resource mapping table using the query key "MOVIE_LIGHTING_MODE". Exact matching requires that the query key be identical to the key value stored in the table, including character case, special symbols, character encoding, and all other details. This strict matching method can avoid query failure or incorrect results due to key value mismatches. The system's query algorithm first calculates the hash value of the query key, then searches for the matching item at the corresponding position in the hash table, and if a hash collision is found, uses a collision resolution strategy to continue searching until a completely matching entry is found or it is confirmed that there is no matching entry. When the system successfully locates an entry in the mapping table that exactly matches the query key, the base resource requirement description information stored in that entry is immediately extracted. The base resource requirement description information is stored in a structured data format, usually using standard formats such as JSON or XML, and contains detailed requirement values for multiple resource types. For example, the extracted resource requirement description may include CPU core count of 2, memory capacity of 4096 MB, GPU computing unit of 0.5, network bandwidth of 10 Mbps, temporary storage space of 2048 MB, and persistent storage space of 512 MB. Each resource type has a clear quantity requirement and performance specification. The system immediately calls a special data parsing module to parse the extracted base resource requirement description. The parsing module first verifies the integrity and correctness of the data format, checks whether it contains all the required resource type information, and whether the value format meets the system requirements. Then the parsing module converts the structured description information into a standard numerical vector format according to the pre-defined resource type order. The conversion process includes unit standardization, data type conversion, numerical precision adjustment, etc. to ensure that the generated vector is compatible with the data format of other system modules. After parsing, the system obtains a standard format base resource requirement vector, and each dimension of the base resource requirement vector corresponds to a specific resource type, with the value representing the amount of that resource required. For example, the final generated base resource requirement vector is [2, 4096, 0.5, 10, 2048, 512], which corresponds to the demand for CPU core number, memory capacity (MB), GPU unit number, network bandwidth (Mbps), temporary storage (MB), and persistent storage (MB) respectively.

[0075] The resource allocation module includes:

[0076] The resource allocation optimization model construction module is configured to establish a resource allocation optimization model with the objective of maximizing the satisfaction of the resource requirements of the candidate target intent nodes and maximizing the resource utilization rate. The basic resource requirement vector is taken as the requirement input constraint of the resource allocation optimization model, and the multi-dimensional available resource vector is taken as the resource supply constraint of the resource allocation optimization model. After obtaining the basic resource requirement vector and the multi-dimensional available resource vector, the system needs to generate an optimal resource allocation scheme through a complex resource scheduling optimization method. The core of the resource scheduling optimization method is to establish a mathematical model that comprehensively considers multiple optimization objectives, that is, to ensure that the resource requirements of the candidate target intent nodes can be fully satisfied and to ensure that the resource utilization rate of the entire system reaches an optimal state. The establishment process of the resource allocation optimization model first needs to determine the specific form of the optimization objective function. The system designs a double-target optimization function, the first target being to maximize the satisfaction degree of the resource requirements of the candidate target intent nodes, and the second target being to maximize the resource utilization rate of the entire system. The resource requirement satisfaction degree is calculated by comparing the actually allocated resource amount with the basic requirement amount, and the satisfaction degree reaches the highest value when the actually allocated amount is equal to or exceeds the basic requirement amount. The system resource utilization rate is measured by calculating the proportion of all allocated resources in the total available resources, which reflects the effective use degree of system resources. Since the two targets may conflict in some cases, because maximizing the requirement satisfaction degree requires that the basic resource requirement vector of the node be satisfied first, which may lead to the fragmentation of resource allocation. Maximizing the resource utilization rate requires that the resource allocation be as compact as possible (such as reducing idle), which may need to sacrifice the requirement satisfaction degree of part of the nodes. The system adopts a weighted summation method to combine them into a single comprehensive objective function, and adjusts the weight coefficient to balance the importance of different targets. The basic resource requirement vector is taken as the requirement input constraint of the model, which specifies that the number of each type of resource allocated to the candidate target intent node cannot be lower than the corresponding basic requirement value, which is the minimum requirement to ensure that the intent analysis task can be normally executed. The multi-dimensional available resource vector is taken as the resource supply constraint of the model, which limits the total amount of each type of resource to be allocated to the available resource amount, which is a reflection of the physical resource limitation of the system.

[0077] The resource allocation scheme generation module is configured to solve the resource allocation optimization model to obtain a candidate allocation scheme set, calculate a resource utilization efficiency evaluation value corresponding to each candidate allocation scheme in the candidate allocation scheme set, and select a candidate allocation scheme with the highest resource utilization efficiency evaluation value as the final resource allocation scheme. The system uses a mixed integer linear programming algorithm to solve the established resource allocation optimization model. The mixed integer linear programming algorithm can handle complex optimization problems containing continuous variables and discrete variables, and is suitable for handling cases where both continuous resources (such as memory capacity and network bandwidth) and discrete resources (such as the number of processor cores and the number of device units) are involved in resource allocation. The mixed integer linear programming algorithm uses a branch and bound method for solving, which approximates the optimal solution by continuously subdividing the problem space and calculating upper and lower bounds, while using pruning strategies to exclude branches that are unlikely to produce optimal solutions, thereby improving the solving efficiency. The solving process generates multiple candidate allocation schemes that satisfy the constraint conditions, and each candidate allocation scheme specifies the number and source of various types of resources allocated to the candidate target intent node in detail. For example, the first candidate scheme provides a minimum configuration that meets the basic needs, allocating 2 CPU cores, 4096 MB of memory, 0.5 GPU units, etc. This scheme is conservative in resource allocation but can guarantee the basic execution needs of the task. The second candidate scheme appropriately increases the resource allocation based on the basic needs, allocating 3 CPU cores, 6144 MB of memory, 1 GPU unit, etc., which can provide better execution performance. The third candidate scheme adopts a relatively aggressive resource allocation strategy, allocating 4 CPU cores, 8192 MB of memory, 1.5 GPU units, etc., which pursues the optimal execution effect but occupies more system resources.

[0078] The resource allocation scheme generation module includes the following steps:

[0079] The resource type and the number of allocated resources in the candidate allocation scheme for the candidate target intent node are obtained to obtain an actual allocation resource vector; the base resource requirement vector corresponding to the candidate target intent node and the current multi-dimensional available resource vector of the resource pool are obtained; in the calculation process of the resource utilization efficiency evaluation value, the system needs to perform comprehensive performance analysis and quantitative evaluation on each candidate allocation scheme to ensure that the selected final scheme can achieve the best balance between meeting the task requirements and optimizing resource utilization. The first step of the calculation process is to obtain the detailed resource information actually allocated to the candidate target intent node in the candidate allocation scheme. For the selected candidate allocation scheme, the system extracts the number of various resource allocations recorded therein to form an actual allocation resource vector. For example, the scheme allocates 3 CPU cores, 6144 MB of memory, 1 GPU computing unit, 15 Mbps of network bandwidth, 3072 MB of temporary storage, and 768 MB of persistent storage to the "movie watching light mode" intent node, and the corresponding actual allocation resource vector is [3, 6144, 1, 15, 3072, 768]. The actual allocation resource vector accurately reflects the specific resource configuration of the allocation scheme. The system simultaneously re-obtains the base resource requirement vector corresponding to the candidate target intent node and the current multi-dimensional available resource vector of the resource pool to ensure that the latest and most accurate data is used in the calculation process. The base resource requirement vector is [2, 4096, 0.5, 10, 2048, 512], which reflects the minimum resource requirement for executing the intent type. The multi-dimensional available resource vector is [8, 16384, 2, 50, 10240, 5120], which represents the available number of various resources in the current resource pool.

[0080] According to the actual allocation resource vector and the basic resource demand vector, a resource demand matching degree is calculated; according to the actual allocation resource vector and the multi-dimensional available resource vector, a resource utilization rate is calculated; the system then calculates a resource demand matching degree index, which reflects the degree of satisfaction of the candidate allocation scheme to the basic resource demand. The calculation method is to compare the actual allocation resource vector and the basic resource demand vector dimension by dimension, for each resource dimension, the ratio of the actual allocation amount to the basic demand amount is calculated, and then the minimum value of the ratio and 1 is taken as the matching degree of the dimension. The CPU core matching degree is min(3 / 2, 1) = 1, indicating that the demand is fully met. The memory matching degree is min(6144 / 4096, 1) = 1, indicating that the demand is fully met. The GPU matching degree is min(1 / 0.5, 1) = 1, indicating that the demand is over-met. The network bandwidth matching degree is min(15 / 10, 1) = 1, indicating that the demand is over-met. The temporary storage matching degree is min(3072 / 2048, 1) = 1, indicating that the demand is over-met. The persistent storage matching degree is min(768 / 512, 1) = 1, indicating that the demand is over-met. The system calculates the overall resource demand matching degree by weighted average calculation of the matching degrees of each dimension, which is 1.0, indicating that the allocation scheme fully meets the basic resource demand. The system then calculates a resource utilization rate index, which reflects the use efficiency of the allocation scheme to the overall resource of the system. The calculation method is to compare the actual allocation resource vector and the multi-dimensional available resource vector, for each resource dimension, the proportion of the actual allocation amount in the total available resource is calculated. The CPU utilization rate is 3 / 8 = 0.375, indicating that 37.5% of the available CPU resource is occupied. The memory utilization rate is 6144 / 16384 = 0.375, indicating that 37.5% of the available memory resource is occupied. The GPU utilization rate is 1 / 2 = 0.5, indicating that 50% of the available GPU resource is occupied. The network bandwidth utilization rate is 15 / 50 = 0.3, indicating that 30% of the available network bandwidth is occupied. The temporary storage utilization rate is 3072 / 10240 = 0.3, indicating that 30% of the available temporary storage is occupied. The persistent storage utilization rate is 768 / 5120 = 0.15, indicating that 15% of the available persistent storage is occupied. The system calculates the resource utilization rate according to the importance weight of each type of resource.

[0081] The resource demand matching degree and the resource utilization rate are weighted to obtain a resource utilization efficiency evaluation value. The system finally weights the resource demand matching degree and the resource utilization rate according to the preset weight proportion to obtain a comprehensive resource utilization efficiency evaluation value of the candidate allocation scheme.

[0082] The resource allocation module further comprises the following steps:

[0083] When the resource allocation scheme determines that the current multi-dimensional available resource vector cannot meet the basic resource requirement vector of the candidate target intent node, all occupied resource intent nodes in the resource pool are scanned, the operation timestamp sequence of the intent node within a predefined time window is extracted, the variance of adjacent operation time intervals is calculated as the behavior dispersion, and the difference between the current time and the last valid operation timestamp of the intent node is obtained as the silence duration; when the system finds that the current multi-dimensional available resource vector cannot fully meet the basic resource requirement of the candidate target intent node through the evaluation of the resource allocation scheme, the pre-emptive resource recycling mechanism needs to be started to release the computing resources occupied by low-activity tasks. All currently occupied resource intent nodes in the resource pool need to be comprehensively scanned and analyzed. The system scanning process traverses all active intent nodes registered in the resource manager, and obtains the basic information of each active intent node, including node identification, intent type, creation time, resource occupation, etc. For each intent node currently occupying resources, the system needs to deeply analyze its recent activity pattern and behavior characteristics to determine whether the intent node is still active or can be safely recycled. The core of the activity analysis is to extract the detailed operation records of each intent node within a predefined time window. The system sets the time window to be 30 minutes in the past, which can capture the short-term behavior pattern of the node and avoid outdated information interference caused by too long window. For each analyzed intent node, the system extracts all operation timestamps of the node within the time window from the operation log to form a timestamp sequence arranged in chronological order. For example, the timestamp sequence of a "background music playing" intent node may be [14:15:30, 14:18:45, 14:22:10, 14:25:55], recording four operation activities in the past 30 minutes. The system then calculates the time intervals between adjacent operation timestamps to evaluate the regularity and continuity of node operation. For the above timestamp sequence, the adjacent intervals are 195 seconds, 205 seconds, and 225 seconds. The system further calculates the statistical variance of these time intervals as the behavior dispersion index of the intent node. The smaller the variance value, the more regular the operation interval, and the more stable the node behavior; the larger the variance value, the more irregular the operation mode, which may indicate a decrease in node activity. Parallel to the behavior dispersion, the silence duration is calculated, which reflects the time length from the last operation of the intent node to the current time. The system obtains the difference between the current system time and the last valid operation timestamp of the node to calculate the silence duration. If the current time is 14:45:30 and the last operation time of the node is 14:25:55, the silence duration is 1175 seconds, about 19.6 minutes. The longer the silence duration, the longer the node has been inactive, and the greater the probability of being judged as inactive.

[0084] The behavior dispersion and the silence duration are used to obtain an active degree score value through a preset inactivity judgment rule, when the active degree score value is lower than a preset active degree score threshold, the intention node is marked as an inactive intention node to be recycled; the inactive intention node is sorted in descending order according to the total amount of resources occupied by the inactive intention node, and the resources of the inactive intention node are recycled into the resource pool in turn. The system uses a preset inactivity judgment rule to comprehensively evaluate the intention node. The preset inactivity judgment rule is a mathematical function, which takes the behavior dispersion and the silence duration as input parameters, and outputs the active degree score value of the intention node. The design of the judgment rule is based on the analysis of a large amount of historical data and the training of a machine learning model, which can accurately reflect the node activity level corresponding to different behavior patterns. The calculation formula of the active degree score value comprehensively considers the time decay effect and the behavior stability factor, for example, active degree score value=(1-silence duration / maximum allowed silence duration)×(1-behavior dispersion / maximum dispersion threshold)×baseline score. After calculation, if the active degree score value of a certain intention node is 0.25, and the preset active degree score threshold is 0.3, it is marked as an inactive intention node to be recycled. The system will add all the inactive intention nodes marked to the recycling candidate list, and then sort them in descending order according to the total amount of resources occupied by the inactive intention nodes. The calculation of the total amount of resources considers the weight and value of different types of resources, and the nodes occupying more high-value resources (such as GPU computing units and large-capacity memories) are preferentially recycled. The execution process of resource recycling includes the coordinated operation of multiple steps. First, a termination signal is sent to the inactive intention node selected for recycling, notifying the relevant processing process to prepare to release the resource. Then the system waits for the process to complete the operation it is currently executing and saves the necessary state information to ensure that data is not lost, and then the system forcibly terminates all associated processes, releases the memory space occupied by them, recycles the processor cores allocated to the node, disconnects the network connection, cleans up temporary files, etc. The system re-labels the recycled resources as available, updates the state information of the resource pool, so that these resources can be used by new high-priority tasks. Through this preemptive resource recycling mechanism, the system can dynamically adjust resource allocation to ensure that important user intention analysis tasks can obtain sufficient computing support, while avoiding the invalid occupation of resources by long-idle tasks, improving the resource utilization efficiency and response performance of the entire system.

[0085] Embodiment 2: Based on the same inventive concept, as shown in Figure 2 The method provided by the embodiment also provides a user intention analysis method based on artificial intelligence, the method comprises the steps of:

[0086] Obtain multi-modal behavior data generated by the user at a current time and a historical time, the multi-modal behavior data including voice data, image data and text data; generate a single-point intent recognition result representing the user's current instantaneous operation according to the multi-modal behavior data at the current time; and generate a time-series intent chain representing the user's continuous operation according to the multi-modal behavior data at the historical time.

[0087] Detect whether there is an intent conflict at present, the intent conflict being a conflict between the single-point intent recognition result and a currently active intent node in the time-series intent chain; when detecting the intent conflict, calculate an optimal transition path from the currently conflicting intent node in the time-series intent chain to a candidate target intent node according to a transition probability weight in a pre-constructed intent transition probability graph.

[0088] Determine resource requirements corresponding to the candidate target intent node of the optimal transition path, and dynamically allocate resources to the candidate target intent node and perform intent analysis according to a current available resource state and the resource requirements.

[0089] Finally, it should be noted that although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments or make equivalent replacements for part of the technical features, and any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An artificial intelligence-based user intent analysis system, characterized by, The system comprises a multi-modal data acquisition and intention generation module, an intention conflict detection and optimal transfer path calculation module, a resource allocation and intention analysis execution module, and sequentially communicates between each module: The multi-modal data acquisition and intention generation module is configured to acquire multi-modal behavior data generated by a user at a current time and historical times, wherein the multi-modal behavior data comprises voice data, image data and text data; generate a single-point intention recognition result representing a current instantaneous operation of the user according to the multi-modal behavior data at the current time; and generate a time-series intention chain representing a continuous operation of the user according to the multi-modal behavior data at the historical times; The intention conflict detection and optimal transfer path calculation module is configured to detect whether an intention conflict exists at the current time, wherein the intention conflict is a conflict between the single-point intention recognition result and a current active intention node in the time-series intention chain; when the intention conflict is detected, calculate an optimal transfer path from a current conflict intention node in the time-series intention chain to a candidate target intention node according to a transfer probability weight in a pre-constructed intention transfer probability graph; The resource allocation and intention analysis execution module is configured to determine a resource requirement corresponding to the candidate target intention node of the optimal transfer path, and dynamically allocate resources to the candidate target intention node and perform intention analysis according to a current available resource state and the resource requirement. 2.The artificial intelligence-based user intent analysis system of claim 1, wherein, The intention conflict detection and optimal transfer path calculation module comprises the following steps: extract semantic feature vectors of the single-point intention recognition result and a semantic feature vector of the current active intention node, and calculate a semantic similarity value between the semantic feature vectors; acquire a duration weight coefficient of the current active intention node in the time-series intention chain; when the semantic similarity value is less than a preset conflict determination threshold, determine that an intention conflict exists; when it is determined that an intention conflict exists, calculate a conflict confidence value according to the semantic similarity value and the duration weight coefficient; and if the conflict confidence value exceeds a preset conflict confidence threshold, confirm the intention conflict. 3.The artificial intelligence-based user intent analysis system of claim 1, characterized in that, The intention conflict detection and optimal transfer path calculation module comprises: a candidate target intention node set determination module configured to acquire a node set in a pre-constructed intention transfer probability graph, a directed edge connecting the nodes, and a transfer probability weight corresponding to each directed edge; take a current conflict intention node in the time-series intention chain as a path starting point, and determine a candidate target intention node set according to a pre-defined screening rule; a maximum cumulative probability and predecessor node analysis module configured to calculate a maximum cumulative probability value and a corresponding predecessor node from the path starting point to each node in the candidate target intention node set according to the transfer probability weight value by using a dynamic programming algorithm; an optimal transfer path calculation module configured to traverse the candidate target intention node set, select a node with the maximum maximum cumulative probability value as an optimal target intention node, and perform reverse backtracking according to a predecessor node corresponding to the optimal target intention node to obtain a complete transfer path from the path starting point to the optimal target intention node as an optimal transfer path. 4.The artificial intelligence-based user intent analysis system of claim 3, characterized in that, The maximum cumulative probability and predecessor node analysis module comprises the following steps: The maximum cumulative probability value corresponding to the path starting point is initialized as a reference value, and the path starting point is marked as having no predecessor node; the maximum cumulative probability values of other nodes are initialized as invalid values; all nodes are traversed in turn according to the topological order of the pre-constructed intention transition probability graph; For the current traversed node, all predecessor nodes corresponding to all incoming edges of the current traversed node are traversed, and the product of the maximum cumulative probability value of the predecessor node and the transition probability weight value of the corresponding incoming edge is calculated as a candidate cumulative probability value; the incoming edge is a directed edge in the intention transition probability graph pointing to the current traversed node; The maximum value of the candidate cumulative probability value is extracted as the maximum cumulative probability value of the current traversed node, and the predecessor node corresponding to the incoming edge generating the maximum cumulative probability value is recorded as the predecessor node of the current traversed node; The above traversal process is repeated to obtain the maximum cumulative probability value and the corresponding predecessor node of each node in the candidate target intention node set. 5.The artificial intelligence-based user intent analysis system of claim 1, characterized in that, The resource allocation and intention analysis execution module comprises: A basic resource demand vector calculation module for retrieving an intention resource mapping table according to the type identifier of the candidate target intention node to obtain a basic resource demand vector; A resource allocation module for scanning the real-time load state of each computing unit in the resource pool to obtain a multi-dimensional available resource vector; obtaining a resource allocation scheme by the resource scheduling optimization method from the basic resource demand vector and the multi-dimensional available resource vector; updating the resource pool after preemptively reclaiming the resources occupied by the non-active intention nodes according to the resource allocation scheme, allocating resources to the candidate target intention nodes according to the updated resource pool; and performing intention analysis on the analysis task after resource allocation is completed. 6.The artificial intelligence-based user intent analysis system of claim 5, wherein, The basic resource demand vector calculation module comprises the following steps: Accessing the intention resource mapping table with the type identifier of the candidate target intention node as a query key, the intention resource mapping table storing key-value pairs with the type identifier of the intention node as the key and the corresponding basic resource demand description as the value; Finding an entry in the intention resource mapping table that exactly matches the query key; extracting the stored basic resource demand description from the matched entry; and obtaining a basic resource demand vector by analyzing the extracted basic resource demand description. 7.The artificial intelligence-based user intent analysis system of claim 5, wherein, The resource allocation module comprises: A resource allocation optimization model construction module for establishing a resource allocation optimization model with the goal of maximizing the satisfaction of the resource demand of the candidate target intention node and maximizing the resource utilization rate; the basic resource demand vector is used as the demand input constraint of the resource allocation optimization model, and the multi-dimensional available resource vector is used as the resource supply constraint of the resource allocation optimization model; A resource allocation scheme generation module for solving the resource allocation optimization model to obtain a candidate allocation scheme set; calculating the resource utilization efficiency evaluation value corresponding to each candidate allocation scheme in the candidate allocation scheme set; and selecting the candidate allocation scheme with the highest resource utilization efficiency evaluation value as the final resource allocation scheme. 8.The artificial intelligence-based user intent analysis system of claim 7, wherein, The resource allocation scheme generation module comprises the following steps: An actual allocation resource vector is obtained by acquiring a resource type and an allocation resource quantity allocated to a candidate target intent node in a candidate allocation scheme; a basic resource requirement vector corresponding to the candidate target intent node and a current multi-dimensional available resource vector of a resource pool are acquired; A resource requirement matching degree is calculated according to the actual allocation resource vector and the basic resource requirement vector; a resource utilization rate is calculated according to the actual allocation resource vector and the multi-dimensional available resource vector; A resource utilization efficiency evaluation value is obtained by weighted calculation of the resource requirement matching degree and the resource utilization rate. 9.The artificial intelligence-based user intent analysis system of claim 5, wherein, The resource allocation module further includes the following steps: When the resource allocation scheme determines that the current multi-dimensional available resource vector cannot meet the basic resource requirement vector of the candidate target intent node, all occupied resource intent nodes in the resource pool are scanned, an operation timestamp sequence of the intent node in a pre-defined time window is extracted, a variance of adjacent operation time intervals is calculated as a behavior dispersion, and a difference between the current time and a last valid operation timestamp of the intent node is obtained as a silence duration; The behavior dispersion and the silence duration are obtained through a pre-set inactivity determination rule to obtain an activity score value, when the activity score value is lower than a pre-set activity score threshold, the intent node is marked as an inactive intent node to be recycled; the total amount of resources occupied by the inactive intent node is sorted in descending order, and the resources of the inactive intent node are recycled into the resource pool in sequence.

10. A user intent analysis method based on artificial intelligence, used to execute the user intent analysis system based on artificial intelligence as described in any one of claims 1-9, characterized in that, The method includes; Multi-modal behavior data generated by a user at a current time and historical times is acquired, the multi-modal behavior data including voice data, image data and text data; a single-point intent recognition result representing current instantaneous operation of the user is generated according to the multi-modal behavior data at the current time; A time-series intent chain representing continuous operation of the user is generated according to the multi-modal behavior data at the historical times; Whether there is an intent conflict is detected, the intent conflict being a conflict between the single-point intent recognition result and a current active intent node in the time-series intent chain; When the intent conflict is detected, an optimal transfer path from a current conflict intent node in the time-series intent chain to a candidate target intent node is calculated according to a transfer probability weight in a pre-constructed intent transfer probability graph; A resource requirement corresponding to the candidate target intent node of the optimal transfer path is determined, resources are dynamically allocated to the candidate target intent node according to a current available resource state and the resource requirement, and intent analysis is performed.

Images