Intention recognition method and related equipment

Abstract

The invention provides an intention recognition method and related equipment, and relates to the technical field of computers. The method comprises the following steps: acquiring multi-dimensional data of a user; performing feature extraction on the multi-dimensional data to obtain multi-dimensional features; performing causal relationship enhancement on the multi-dimensional features and at least one preset intention to obtain causal features; and processing the multi-dimensional features and the causal features through the intention recognition model to obtain a target intention recognition result, the intention recognition interpretability is enhanced and improved through the causal relationship, the recognition accuracy and robustness are effectively improved, and the user experience is effectively improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to an intent recognition method, an intent recognition device, an electronic device, a computer-readable storage medium, and a computer program product. Background Art

[0002] User intent recognition on carriers' online platforms is a crucial component of modern telecommunications services. This technology can help carriers automatically understand user requests and provide feedback, reducing the workload of human customer service representatives.

[0003] In related technologies, user intent recognition mainly relies on understanding user input sentences. However, this approach has the problem that the prediction results are easily affected by confusion factors and lack accuracy and robustness.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute prior art known to ordinary technicians in the field. Summary of the Invention

[0005] The purpose of the present disclosure is to provide an intent recognition method and related equipment, which can at least to some extent overcome the problems in the related art that the prediction results of intent recognition methods are easily affected by confusion factors and lack accuracy and robustness.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by practice of the present disclosure.

[0007] According to one aspect of the present disclosure, a method for intent recognition is provided, including: acquiring multidimensional data of a user; performing feature extraction on the multidimensional data to obtain multidimensional features; performing causal enhancement on the multidimensional features and at least one preset intent to obtain causal features; and processing the multidimensional features and the causal features through an intent recognition model to obtain a target intent recognition result.

[0008] In one embodiment of the present disclosure, the causal relationship enhancement of the multidimensional features and at least one preset intention to obtain the causal feature includes: constructing a completely undirected graph based on the multidimensional features and the at least one preset intention, wherein each node in the completely undirected graph represents a feature or an intention, and an edge between two nodes in the completely undirected graph indicates that there may be a dependency relationship between the two nodes; performing a conditional independence test on each pair of nodes in the completely undirected graph to construct a local undirected graph, wherein a pair of nodes connected by an edge in the local undirected graph are nodes that have a dependency relationship under the condition of given other nodes; determining the direction of the edge in the local undirected graph to construct a causal graph; determining the weight corresponding to each feature node in the causal graph to obtain the causal feature.

[0009] In one embodiment of the present disclosure, the conditional independence test is performed on each pair of nodes in the completely undirected graph to construct a local undirected graph, including: given the other nodes in the completely undirected graph, if a pair of nodes in the completely undirected graph is independent, then the edge between the pair of nodes is deleted, and the pair of nodes includes a feature node and an intention node.

[0010] In one embodiment of the present disclosure, if a pair of nodes in the completely undirected graph are independent given other nodes in the completely undirected graph, then the edge between the pair of nodes is deleted, including: calculating the chi-square value of the one feature node and the one intention node given other nodes in the completely undirected graph; if the calculated chi-square value is less than or equal to a preset quantity threshold, then determining that the one feature node and the one intention node are independent.

[0011] In one embodiment of the present disclosure, determining the direction of the edge in the local undirected graph and constructing a causal graph includes: using a preset verification method to determine the direction of the edge in the undirected graph.

[0012] In one embodiment of the present disclosure, determining the weights corresponding to the characteristic nodes in the causal graph to obtain the causal features includes: in a path from a characteristic node to a path node, if the path is within the second order, determining the weights of the characteristic nodes in the path as a first value; if it is a path from an intention node to a characteristic node or the path reaches the third order or above, determining the weights of the characteristic nodes in the path as a second value, wherein the second value is less than the first value, and the second value is determined by a random number generation method.

[0013] In one embodiment of the present disclosure, the intent recognition model includes a first conversion layer and a second conversion layer; wherein, the multidimensional features and the causal features are processed by the intent recognition model to obtain a target intent recognition result, including: performing semantic recognition on the multidimensional features through the first conversion layer to obtain a semantic vector; processing the causal features and the semantic vector through the second conversion layer to obtain a probability value of each preset intention; determining the preset intention with the largest probability value as the target intent recognition result; wherein, the intent recognition model is trained by the user's multidimensional historical data and at least one preset intention label.

[0014] In one embodiment of the present disclosure, after the multidimensional features and the causal features are processed by the intent recognition model to obtain a target intent recognition result, the method includes: determining a target execution module corresponding to the target intent recognition result, and diverting the user to the target execution module.

[0015] According to another aspect of the present disclosure, an intention recognition device is provided, including: a data acquisition module for acquiring multidimensional data of a user; a feature extraction module for performing feature extraction on the multidimensional data to obtain multidimensional features; a feature enhancement module for performing causal enhancement on the multidimensional features and at least one preset intention to obtain causal features; and an intention recognition module for processing the multidimensional features and the causal features through an intention recognition model to obtain a target intention recognition result.

[0016] According to another aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor; the processor is configured to perform the above-mentioned intention recognition method by executing the executable instructions.

[0017] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned intention recognition method is implemented.

[0018] According to another aspect of the present disclosure, a computer program product is provided, on which a computer program is stored. When the computer program is executed by a processor, the above-mentioned intention recognition method is implemented.

[0019] In the implementation manner of the present disclosure, multidimensional data of the user is obtained; feature extraction is performed on the multidimensional data to obtain multidimensional features; causal relationship enhancement is performed on the multidimensional features and at least one preset intention to obtain causal features; the multidimensional features and causal features are processed by an intent recognition model to obtain a target intent recognition result. The present disclosure improves the interpretability of intent recognition through causal relationship enhancement, effectively improves recognition accuracy and robustness, and effectively improves user experience.

[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and those skilled in the art can derive other drawings based on these drawings without inventive effort.

[0022] Figure 1 A schematic diagram illustrating an exemplary application system architecture of an intent recognition method in an embodiment of the present disclosure is shown.

[0023] Figure 2 A flow chart of an intent recognition method in an embodiment of the present disclosure is shown.

[0024] Figure 3 A flow chart of a method for determining causal characteristics in an embodiment of the present disclosure is shown.

[0025] Figure 4 A flow chart of another method for determining causal characteristics in an embodiment of the present disclosure is shown.

[0026] Figure 5 A flow chart of another method for determining causal characteristics in an embodiment of the present disclosure is shown.

[0027] Figure 6 A flow chart of a method for determining target intention recognition results in an embodiment of the present disclosure is shown.

[0028] Figure 7 A structural diagram of an intent recognition model in an embodiment of the present disclosure is shown.

[0029] Figure 8 A flowchart of another intent recognition method in an embodiment of the present disclosure is shown.

[0030] Figure 9 An example flowchart of an intent recognition method in an embodiment of the present disclosure is shown.

[0031] Figure 10 A structural schematic diagram of an example of a completely undirected graph in an embodiment of the present disclosure is shown.

[0032] Figure 11 A schematic structural diagram of an example of a cause-effect diagram in an embodiment of the present disclosure is shown.

[0033] Figure 12 A schematic structural diagram of an intention recognition device in an embodiment of the present disclosure is shown.

[0034] Figure 13 A schematic structural diagram of an example of an intention recognition device in an embodiment of the present disclosure is shown.

[0035] Figure 14 A structural block diagram of an electronic device in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0036] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be embodied in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0037] In addition, the accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0038] Figure 1 FIG. 1 shows an exemplary application system architecture diagram to which the intention recognition method in the embodiment of the present disclosure can be applied. Figure 1 As shown, the system architecture may include a terminal device 101 , a network 102 and a server 103 .

[0039] The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103 , and can be a wired network or a wireless network.

[0040] Optionally, the above-mentioned wireless network or wired network uses standard communication technologies and / or protocols. The network is typically the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network or any combination of a virtual private network). In some embodiments, technologies and / or formats including Hypertext Markup Language (HTML), Extensible Markup Language (XML), etc. are used to represent data exchanged over the network. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPSec) can be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above-mentioned data communication technologies.

[0041] The terminal device 101 can be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, smart speakers, smart watches, wearable devices, augmented reality devices, virtual reality devices, etc.

[0042] Optionally, the client of the application installed in different terminal devices 101 is the same, or the client of the same type of application based on different operating systems. The specific form of the application client can also be different based on the terminal platform. For example, the application client can be a mobile phone client, a PC client, etc. The application installed on the terminal device 101 can collect interaction information between the user and the platform customer service or intelligent assistant.

[0043] The server 103 may be a server that provides various services, such as a background management server that provides support for the devices operated by the user using the terminal device 101. The background management server may analyze and process the collected multidimensional data and other data, and feed back the processing results to the terminal device 101.

[0044] Optionally, the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0045] Those skilled in the art will know that Figure 1 The number of terminal devices, networks, and servers in the embodiment is merely illustrative, and any number of terminal devices, networks, and servers may be provided based on actual needs. This embodiment of the present disclosure does not limit this.

[0046] User intent recognition on carrier online platforms is a crucial component of modern telecommunications services. User intent recognition technology can help carriers automatically understand customer requests and provide accurate responses, reducing the workload of customer service representatives and improving the user experience. Users often encounter issues such as incorrect input and difficulty completing transactions accurately and directly when processing carrier services. Products on carrier platforms are highly homogeneous, with hundreds of similar packages available. Therefore, it is crucial to accurately identify potential user intent based on user descriptions and historical behavior.

[0047] Related technologies rely on understanding user input to identify user intent. However, this approach ignores user behavior and historical information when logging into the platform. Existing algorithms also fail to fully consider the causal relationship between features and user intent. This makes telecom user intent prediction susceptible to confounding factors, resulting in insufficient accuracy and robustness.

[0048] In order to at least solve the above technical problems, the intention recognition method provided by the embodiment of the present disclosure obtains multidimensional data of the user; performs feature extraction on the multidimensional data to obtain multidimensional features; performs causal enhancement on the multidimensional features and at least one preset intention to obtain causal features; and processes the multidimensional features and causal features through the intention recognition model to obtain the target intent recognition result. The present disclosure improves the interpretability of intent recognition through causal enhancement, effectively improves recognition accuracy and robustness, and effectively improves user experience.

[0049] Under the above system architecture, the embodiment of the present disclosure provides an intention recognition method, which can be executed by any electronic device with computing and processing capabilities. In some embodiments, the intention recognition method provided in the embodiment of the present disclosure can be executed by a server of the above system architecture.

[0050] The intention recognition method in this example implementation will be described in more detail below with reference to the accompanying drawings and examples.

[0051] Figure 2 FIG. 1 shows a flow chart of an intention recognition method according to an embodiment of the present disclosure. Figure 2 As shown, in one embodiment, the intention recognition method disclosed in the present invention mainly includes the following steps:

[0052] S202: Acquire multi-dimensional data of the user.

[0053] In one embodiment, the dimensions of the multidimensional data may include but are not limited to user behavior dimension, user-platform interaction dimension, platform dimension, etc.

[0054] Multidimensional data can include data from at least two dimensions: user behavior data, user interaction data, and platform marketing data. User behavior data can include user behavior records on the platform, such as browsing, clicking, and purchasing records. User interaction data includes conversations between users and the platform's customer service or intelligent assistant, such as multiple rounds of conversations between users and platform customer service. User interaction data can be text interaction information or text interaction information converted from voice interaction information. Platform marketing data can include platform marketing activity data, such as advertising and promotional activities.

[0055] In one embodiment, after obtaining the user's multidimensional data, the multidimensional data can also be cleaned and preprocessed. For example, invalid data or erroneous data can be removed from the multidimensional data, and then the cleaned multidimensional data can be formatted and standardized and other preprocessing operations can be performed to obtain multidimensional data that can be used for model training or model prediction.

[0056] S204: Extract features from the multidimensional data to obtain multidimensional features.

[0057] In one embodiment, multidimensional features may be extracted from multidimensional data. The multidimensional features may include user behavior features, user interaction features, and context features.

[0058] User interaction features can be extracted from user interaction data, focusing on multiple rounds of conversations between users and the platform's customers or intelligent assistants. Natural language processing technology can be used to perform pre-processing operations such as word segmentation, part-of-speech tagging, and sentiment analysis on the text data corresponding to the multiple rounds of conversations to extract key information and quantify it to obtain user interaction features.

[0059] User behavior characteristics can be extracted from user behavior data. By analyzing users' browsing paths, page dwell time, click frequency, etc., statistical data of various behaviors can be obtained, and user behavior characteristics can be constructed based on the statistical data.

[0060] Contextual features can be extracted from platform marketing data. For example, information such as the user's login time, location, device type, promotions and discounts on the platform page can be extracted, and the extracted key information can be vectorized to obtain contextual features.

[0061] S206: Perform causal relationship enhancement on the multi-dimensional features and at least one preset intention to obtain causal features.

[0062] In one embodiment, the purpose of causal enhancement is to filter and enhance features that help identify user intent, that is, to filter out features that are more closely related to the preset intent, thereby improving the accuracy of intent recognition.

[0063] The number of the above-mentioned preset intentions can be determined according to the number of function execution modules actually set by the platform, and this disclosure does not limit this. For example, the preset intentions may include purchasing traffic, changing packages, after-sales consultation, business appointments, etc.

[0064] In one embodiment, the Peter-Clark algorithm (PC algorithm) can be used to infer the causal relationship between the feature and the preset intention to obtain a directed acyclic graph. In the directed acyclic graph, the direction and weight indicate the causal relationship between the feature and the intention. The causal relationship can be used to enhance the multidimensional features to obtain causal features.

[0065] S208. Process the multidimensional features and causal features through the intention recognition model to obtain the target intention recognition result.

[0066] In one embodiment, the intent recognition model represents the mapping relationship between multi-dimensional features, causal features, and target intent recognition results. The intent recognition model can be implemented based on a hierarchical transformer.

[0067] Through multiple Transformer layers, we can gradually extract and analyze the intention information in multidimensional features and causal features, and then determine the user's intention.

[0068] The target intent recognition result may be one of at least one preset intent, or may not be a preset intent. For example, none of the above preset intents can meet the user's needs.

[0069] In an embodiment of the present disclosure, multidimensional data of the user is obtained; feature extraction is performed on the multidimensional data to obtain multidimensional features; causal relationship enhancement is performed on the multidimensional features and at least one preset intention to obtain causal features; the multidimensional features and causal features are processed by an intent recognition model to obtain a target intent recognition result. The present disclosure improves the interpretability of intent recognition through causal relationship enhancement, effectively improves recognition accuracy and robustness, and effectively improves user experience.

[0070] Figure 3 A flow chart of a method for determining causal characteristics in an embodiment of the present disclosure is shown. Figure 3 As shown, in one embodiment, S206 performs causal relationship enhancement on the multi-dimensional features and at least one preset intention to obtain causal features, including:

[0071] S302: Construct a completely undirected graph based on the multidimensional features and at least one preset intent, where each node in the completely undirected graph represents a feature or an intent, and an edge between two nodes in the completely undirected graph represents a possible dependency relationship between the two nodes;

[0072] S304, performing a conditional independence test on each pair of nodes in the completely undirected graph, and constructing a local undirected graph, wherein a pair of nodes connected by an edge in the local undirected graph is a node that has a dependency relationship given other nodes;

[0073] S306, determining the direction of the edge in the local undirected graph and constructing a causal graph;

[0074] S308: Determine the weight corresponding to each feature node in the causal graph to obtain causal features.

[0075] In S302 , a feature corresponds to a node in the completely undirected graph, and a preset intent corresponds to a node in the completely undirected graph, that is, a node in the completely undirected graph can be a feature node or an intent node.

[0076] For any node, you can connect it to other nodes through edges, forming a completely undirected graph. For example, a multidimensional feature consists of 6 features and 2 preset intents. In a completely undirected graph, there are 8 nodes: 6 feature nodes (X1, X2, X3, X4, X5, X6) and 2 intent nodes (Y1, Y2). Feature node X1 is connected to feature nodes X2, X3, X4, X5, X6, and intent nodes Y1 and Y2 by an edge, and so on. In a completely undirected graph, a total of C(8,2) = 28 edges are required.

[0077] The two nodes connected by an edge in a completely undirected graph can be two feature nodes or a feature node and an intent node.

[0078] In S304 , the purpose of the conditional independence test is to test the independence of two nodes connected by an edge. When the two nodes connected by the edge are independent, the edge is deleted from the completely undirected graph.

[0079] In a local undirected graph, any pair of nodes are nodes that have a dependency relationship given other nodes, that is, two nodes that do not have an edge connection in a local undirected graph are independent.

[0080] It should be noted that the independence of the two nodes connected by each edge in a completely undirected graph needs to be determined.

[0081] In S306, in the causal graph, the direction of the edge is from the cause node to the effect node. For example, the direction of the edge between the feature node X1 and the intention node Y1 is from the feature node X1 to the intention node Y1, indicating that the feature node X1 is a possible cause of the intention node Y1.

[0082] It should be noted that the directions of all edges in the causal graph need to be determined.

[0083] In one embodiment, the above S306 determines the direction of the edge in the local undirected graph and constructs the causal graph, including: using a preset verification method to determine the direction of the edge in the undirected graph.

[0084] The preset test method may be a d separation test, a Fisher Z Test, or the like, and is not specifically limited thereto.

[0085] In S308, in the causal graph, the paths from the feature nodes to the intention nodes can be counted, and the number of nodes and node types in each path can be recorded. The weight of each path can be determined based on the number of nodes and node types, and the weight of each path can be used as a causal feature to enhance the causal relationship.

[0086] In an embodiment of the present disclosure, a completely undirected graph is constructed based on multidimensional features and at least one preset intention, a conditional independence test is performed on each pair of nodes in the completely undirected graph, a local undirected graph is constructed, the direction of the edges in the local undirected graph is determined, a causal graph is constructed, the weights corresponding to each feature node in the causal graph are determined, and causal features are obtained. The multidimensional features can be enhanced based on the causal relationship, thereby screening out features that are beneficial for intent recognition, greatly improving the accuracy of user intent recognition.

[0087] It should be noted that, in addition to using the PC algorithm to construct a causal graph and determine causal features, a scoring-based algorithm may also be used, such as the Greedy Equivalence Search (GES) algorithm.

[0088] Figure 4 A flow chart of another method for determining causal characteristics in an embodiment of the present disclosure is shown. Figure 3 Based on the embodiment, S304 is further refined into S3042 to define the process of constructing a local undirected graph. Figure 4 As shown, in one embodiment, the above S304 performs a conditional independence test on each pair of nodes in the completely undirected graph to construct a local undirected graph, including:

[0089] S3042. Given other nodes in a completely undirected graph, if a pair of nodes in the completely undirected graph are independent, delete the edge between the pair of nodes, where the pair of nodes includes a feature node and an intention node.

[0090] It should be noted that, assuming that multiple preset intents are independent, after constructing a completely undirected graph, the edges connecting intent nodes in the completely undirected graph can be deleted, or the edges connecting intent nodes can be included in the constructed completely undirected graph.

[0091] A pair of nodes in a completely undirected graph is independent, which means that a feature node and an intent node are independent.

[0092] In one embodiment, the above-mentioned method of deleting the edge between a pair of nodes in a completely undirected graph is independent under the condition of other nodes in the completely undirected graph, including: calculating the chi-square value of a feature node and an intention node under the condition of other nodes in the completely undirected graph; if the calculated chi-square value is less than or equal to a preset quantity threshold, determining that a feature node and an intention node are independent.

[0093] In this disclosure, the chi-square test is used to determine whether the feature node and the intention node are independent. 2 It can be calculated using the following formula:

[0094]

[0095] Among them, O represents the observed frequency and E represents the expected frequency.

[0096] When the chi-square value χ 2 The closer it is to 0, the weaker the correlation between the feature node and the intention node, and the closer they are to being completely independent. In this case, the edge connecting the feature node and the intention node is deleted. 2 When the number is less than or equal to the preset threshold, it is determined that the feature node has a certain association with the intention node, but the feature is not sufficient to lead to the intention. At this time, the edge connecting the feature node and the intention node is deleted. 2 When the number is greater than a preset threshold, it is determined that the feature node has a strong correlation with the intention node. At this time, the edge connecting the feature node and the intention node is retained.

[0097] The above-mentioned preset quantity threshold can be determined according to actual needs, and this disclosure does not make any specific limitation on this. For example, the preset quantity threshold is 6.635.

[0098] In the embodiment of the present disclosure, by deleting the edges corresponding to two independent nodes, two nodes that do not have a causal relationship can be eliminated, thereby enhancing the causal relationship, improving the accuracy of user intention prediction, and greatly improving user experience.

[0099] Figure 5 A flow chart of another method for determining causal characteristics in an embodiment of the present disclosure is shown. Figure 5 As shown, in one embodiment, the above S308 determines the weight corresponding to each feature node in the causal graph to obtain the causal feature, including:

[0100] S3082. In a path from a feature node to a path node, if the path is within the second order, the weight of the feature node in the path is determined to be a first value; if it is a path from an intention node to a feature node or the path reaches the third order or above, the weight of the feature node in the path is determined to be a second value, wherein the second value is smaller than the first value, and the second value is determined by random number generation.

[0101] In one embodiment, in a causal graph, for each intention node, there may be multiple paths from feature nodes to the intention node.

[0102] A path can include one or more feature nodes. The order of the path is determined by the number of feature nodes, with the order of feature nodes decreasing as the path moves toward the intended node. For example, a path with two feature nodes is called a second-order path, while a path with three feature nodes is called a third-order path.

[0103] The first value may be a fixed value or a group of fixed values.

[0104] For a first-order path, the weight of the feature node in the first-order path can be determined as 1, that is, the first value is 1; for a second-order path, the weight of the feature node directly pointing to the intention node can be determined as 1, and the weight of the feature node indirectly pointing to the intention node can be determined as 0.8, which is used as the final feature weight value, that is, the first value of the first-order feature node is 1, and the first value of the second-order feature node is 0.8.

[0105] For paths of third order or above, or paths where the intent node points to the feature node, the second value can be between 0 and 0.5, and the second value can be generated by a random number generation method.

[0106] It should be noted that the farther away the feature node is from the intention node, the larger the order of the feature node is and the smaller its weight is; the closer the feature node is to the intention node, the smaller the order of the feature node is and the larger its weight is.

[0107] In the embodiments of the present disclosure, the weights of the feature nodes in the path are determined according to the end of the path, thereby screening out features that are beneficial to user intent recognition, improving the accuracy of user intent prediction, and greatly improving user experience.

[0108] Figure 6 FIG. 1 is a flow chart of a method for determining a target intention recognition result according to an embodiment of the present disclosure. Figure 6 As shown, in one embodiment, the intent recognition model includes a first conversion layer and a second conversion layer; wherein, the above S208 processes the multidimensional features and causal features through the intent recognition model to obtain the target intent recognition result, including:

[0109] S602, performing semantic recognition on the multi-dimensional features through the first conversion layer to obtain a semantic vector;

[0110] S604: Process the causal features and semantic vectors through the second conversion layer to obtain probability values ​​of each preset intention;

[0111] S606. Determine the preset intention with the largest probability value as the target intention recognition result; wherein the intention recognition model is obtained by training the user's multi-dimensional historical data and at least one preset intention label.

[0112] In one embodiment, the first transformation layer and the second transformation layer may be composed of a complete Transformer module.

[0113] In S602, the multidimensional features are input to the first transformation layer, which performs semantic recognition and outputs a semantic vector containing rich semantic information. The first transformation layer can focus on extracting semantic information from multidimensional features. The Transformer module uses a self-attention mechanism to capture long-range dependencies between features, thereby better understanding the semantic content of the features.

[0114] In S604, the semantic vector and causal features are input to the second conversion layer, which processes the semantic vector and outputs the probability value of each preset intention. The second conversion layer further analyzes the semantic information of the semantic vector to determine the user's intention.

[0115] Figure 7 FIG. 1 is a schematic diagram showing a structure of an intention recognition model in an embodiment of the present disclosure. Figure 7 As shown, the multidimensional features are input into the first conversion layer and the semantic vector is output; the causal features are determined based on the multidimensional features and the causal graph; the semantic vector and the causal features are input into the second conversion layer, the probability value of the preset intention is output, and the preset intention with the largest probability value is determined as the target intention recognition result.

[0116] This can be achieved through the Transformer architecture, which includes a multi-head attention mechanism to capture intent-related features from different perspectives. In the second transformation layer, the self-attention mechanism continues to play a role, further identifying and strengthening the features most relevant to user intent. Through this approach, the intent recognition model can extract the information that best represents user intent from the output of the first transformation layer. After passing it through a softmax layer, it obtains the probability of user intent and makes a prediction.

[0117] A dataset can be constructed based on multidimensional historical data and at least one preset intent label. The dataset can be divided into a training set and a validation set according to a preset ratio (e.g., 8:2, 7:3, etc.). The intent recognition model to be trained is trained using the training set to obtain an initial intent recognition model. The initial intent recognition model is then validated using the validation set to obtain a final intent recognition model. There are no restrictions on the training process of the intent recognition model.

[0118] In the disclosed embodiment, a hierarchical intent recognition model based on Transformer gradually refines and strengthens causal inference and deep learning technology through hierarchical structural design. The model can more effectively handle complex user intent recognition tasks.

[0119] Figure 8 Another flow chart of the intention recognition method in the embodiment of the present disclosure is shown. Figure 1 On the basis of the embodiment, S210 is added after S208 to limit the solution of making system decisions based on the target intention recognition results. Figure 8 As shown, the intention recognition method of the present disclosure includes S202 to S210. In S208, after the multidimensional features and causal features are processed by the intention recognition model to obtain the target intention recognition result, the method includes:

[0120] S210: Determine a target execution module corresponding to the target intention recognition result, and divert the user to the target execution module.

[0121] The target execution module is a functional module corresponding to the target intent recognition result. The target execution module can include purchasing traffic, changing packages, after-sales consultation, business appointments, etc. The target intent recognition result determines what actions the platform will take next, thereby optimizing user services.

[0122] After the intention recognition model training is completed, when the user enters the platform, the causal relationship is enhanced according to the current action to obtain the causal features. The causal features and multi-dimensional features are input into the intention recognition model to realize the user's personalized dynamic intention recognition. The system jumps to the corresponding functional module according to the recognition result of the target intent to realize the subsequent user's required operations.

[0123] In order to deepen the understanding of the intention recognition method disclosed in this disclosure, the following Figures 9 to 11 For detailed description. Figure 9 As shown, an embodiment of the present disclosure provides an intent recognition method, including:

[0124] S901, data collection;

[0125] The collected multi-dimensional data may include at least two of the following: user behavior data, user interaction data, and platform marketing data. User behavior data may include user behavior records on the platform. User interaction data may include conversations between users and the platform's customer service or intelligent assistant. Platform marketing data may include data on the platform's marketing activities, such as advertising and promotional activities.

[0126] S902, data cleaning and preprocessing;

[0127] Invalid or erroneous data is removed from the multidimensional data, and then preprocessing operations such as formatting and standardization are performed on the cleaned multidimensional data to obtain multidimensional data that can be used for model training or model prediction.

[0128] S903, feature extraction;

[0129] Multidimensional features can be extracted from multidimensional data. The multidimensional features may include user behavior features, user interaction features, and context features.

[0130] S904, constructing a cause-effect diagram;

[0131] Through the causal graph, the causal relationship of multidimensional features can be enhanced to obtain causal features.

[0132] S905: training a hierarchical intent recognition model;

[0133] A data set is constructed based on multidimensional features, causal features, and user intention labels. A hierarchical intent recognition model is trained based on the data set to obtain a trained intent recognition model.

[0134] S906: Predict user intent based on the trained intent recognition model.

[0135] The collected real-time multi-dimensional user input is input into the trained intent recognition model to predict user intent.

[0136] Figure 10 FIG. 1 is a schematic diagram showing a completely undirected graph example in an embodiment of the present disclosure. Figure 10As shown, the completely undirected graph includes 5 feature nodes and 2 intention nodes. The 5 feature nodes are click: package page, age 25, remaining traffic 0G, browse: directed traffic, and other features, and the 2 intention nodes are purchase traffic and after-sales consultation.

[0137] The number of edges in a completely undirected graph is strip.

[0138] Figure 11 A schematic structural diagram of an example of a cause-effect diagram in an embodiment of the present disclosure is shown. Figure 11 for Figure 10 The causal diagram obtained after causal discovery is Figure 11 For the intention node of purchasing traffic, there are two paths in total. Path 1 is age 25-remaining traffic 0G-purchase traffic, and path 2 is path 1 is age 25-remaining traffic 0G-browsing: targeted traffic-purchase traffic; for the intention node of after-sales consultation, there is one path: click: package page-other nodes-after-sales consultation.

[0139] Based on the same inventive concept, the present disclosure also provides an intention recognition device, as described in the following embodiments. Since the principle of solving the problem in the device embodiment is similar to that in the above method embodiment, the implementation of the device embodiment can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0140] Figure 12 FIG. 1 is a schematic diagram showing the structure of an intention recognition device according to an embodiment of the present disclosure. Figure 12 As shown, in one embodiment, the device includes: a data acquisition module 1210, used to obtain multidimensional data of the user; a feature extraction module 1220, used to extract features from the multidimensional data to obtain multidimensional features; a feature enhancement module 1230, used to enhance the causal relationship between the multidimensional features and at least one preset intention to obtain causal features; an intention recognition module 1240, used to process the multidimensional features and causal features through an intention recognition model to obtain a target intention recognition result.

[0141] It should be noted that the data acquisition module 1210, feature extraction module 1220, feature enhancement module 1230, and intent recognition module 1240 described above correspond to S202 to S208 in the method embodiment. The examples and application scenarios implemented by these modules and corresponding steps are the same, but are not limited to the contents disclosed in the method embodiment described above. It should be noted that the above modules, as part of the apparatus, can be executed in a computer system, such as a set of computer-executable instructions.

[0142] In one embodiment, the feature enhancement module 1230 is used to construct a completely undirected graph based on multidimensional features and at least one preset intent, where each node in the completely undirected graph represents a feature or an intent, and an edge between two nodes in the completely undirected graph indicates that there may be a dependency relationship between the two nodes; a conditional independence test is performed on each pair of nodes in the completely undirected graph to construct a local undirected graph, where a pair of nodes connected by an edge in the local undirected graph are nodes that have a dependency relationship under the condition of given other nodes; the direction of the edge in the local undirected graph is determined to construct a causal graph; the weights corresponding to each feature node in the causal graph are determined to obtain causal features.

[0143] In one embodiment, the feature enhancement module 1230 is further used to delete the edge between a pair of nodes in a completely undirected graph, given other nodes in the completely undirected graph, if the pair of nodes are independent, and the pair of nodes includes a feature node and an intention node.

[0144] In one embodiment, the feature enhancement module 1230 is also used to calculate the chi-square value of a feature node and an intention node given other nodes in a completely undirected graph; if the calculated chi-square value is less than or equal to a preset quantity threshold, it is determined that a feature node and an intention node are independent.

[0145] In one embodiment, the feature enhancement module 1230 is further configured to use a preset verification method to determine the direction of an edge in an undirected graph.

[0146] In one embodiment, the feature enhancement module 1230 is also used to determine the weight of the feature node in the path as a first value if the path is within the second order in a path from a feature node to a path node; if the path is from an intention node to a feature node or the path reaches the third order or above, the weight of the feature node in the path is determined as a second value, wherein the second value is smaller than the first value and the second value is determined by random number generation.

[0147] In one embodiment, the intent recognition model includes a first conversion layer and a second conversion layer; the intent recognition module 1240 is used to perform semantic recognition on multidimensional features through the first conversion layer to obtain a semantic vector; the causal features and the semantic vector are processed through the second conversion layer to obtain the probability value of each preset intention; the preset intention with the largest probability value is determined as the target intent recognition result; wherein the intent recognition model is trained through the user's multidimensional historical data and at least one preset intention label.

[0148] In one embodiment, the device also includes a display module not shown in the accompanying drawings, and the display module is used to process the multidimensional features and causal features through the intention recognition model to obtain the target intention recognition result, determine the target execution module corresponding to the target intention recognition result, and divert the user to the target execution module.

[0149] In the implementation manner of the present disclosure, multidimensional data of the user is obtained; feature extraction is performed on the multidimensional data to obtain multidimensional features; causal relationship enhancement is performed on the multidimensional features and at least one preset intention to obtain causal features; the multidimensional features and causal features are processed by an intent recognition model to obtain a target intent recognition result. The present disclosure improves the interpretability of intent recognition through causal relationship enhancement, effectively improves recognition accuracy and robustness, and effectively improves user experience.

[0150] In order to deepen the understanding of the intention recognition device disclosed in the present invention, the following Figure 13 Provide detailed explanation.

[0151] like Figure 13 As shown, the intention recognition device disclosed in the present invention includes a data acquisition module, a feature engineering module, an intention recognition module, a function execution module and a data storage module.

[0152] The data storage module is used for data storage and update, etc.

[0153] The data collection module can collect user behavior data, user conversation data, and platform marketing data to obtain multidimensional data, which can be stored in the data storage module. User behavior data can include user behavior records on the platform, such as browsing, clicking, and purchasing; user conversation data can include conversations between users and platform clients or intelligent assistants, such as interactive text messages and voice messages; and platform marketing data can include platform marketing activity data, such as advertising and promotional activities.

[0154] The feature engineering module includes the aforementioned feature extraction module 1220 and feature enhancement module 1230. The feature engineering module is used for data cleaning and preprocessing, constructing feature sets, building causal graphs, and selecting and enhancing features. The feature engineering module cleans the multidimensional data collected by the data acquisition module, removes invalid or erroneous data, and formats and standardizes it. The feature extraction module 1220 then extracts vectorized features from the processed multidimensional data. The feature enhancement module 1230 explores the causal relationship between the vectorized features and pre-set intent, constructing a causal graph to screen and enhance features that help identify user intent.

[0155] The intent recognition module is responsible for model construction, model training, model inference, and platform action decision-making. Based on the processed data and optimized features, the intent recognition module constructs a hierarchical Transformer model for intent recognition, generating a trained intent recognition model. The trained intent recognition model is trained using the training dataset, and the optimal model is selected using the test dataset, resulting in the trained intent recognition model. The trained intent recognition model is then used to identify user intent using multi-dimensional data, generating target intent recognition results and accurately identifying user intent. Based on the target intent recognition results, the platform determines actions, such as service or product recommendations.

[0156] The function execution module can include purchasing traffic, changing packages, after-sales consultation, business appointments, etc., and decides what actions the platform should take based on user intentions, thereby optimizing user services.

[0157] The data storage module is responsible for storing collected multidimensional data, extracted multidimensional features, training parameters, and regularly updated data. Collected multidimensional data, extracted multidimensional features, causal features, trained model parameters, and user feedback are all stored in the data storage module and regularly updated to maintain data timeliness and accuracy.

[0158] In this disclosure, through the collaborative work between the above modules, a complete process from data collection to intent recognition to decision execution can be achieved, thereby improving user experience and platform operation efficiency.

[0159] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods, or program products. Therefore, various aspects of the present disclosure may be implemented in the following forms: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software implementations, which may be collectively referred to herein as "circuits," "modules," or "systems."

[0160] Refer to the following Figure 14 1400 according to this embodiment of the present disclosure will be described. Figure 14 The electronic device 1400 shown is merely an example and should not limit the functions and scope of use of the embodiments of the present disclosure.

[0161] like Figure 14 As shown, electronic device 1400 is implemented as a general-purpose computing device. Components of electronic device 1400 may include, but are not limited to, the aforementioned at least one processing unit 1410, the aforementioned at least one storage unit 1420, and a bus 1430 connecting various system components (including storage unit 1420 and processing unit 1410).

[0162] The storage unit stores program code, which can be executed by the processing unit 1410, so that the processing unit 1410 performs the steps described in the "Exemplary Method" section of this specification according to various exemplary embodiments of the present disclosure. For example, the processing unit 1410 can perform the following steps of the above method embodiment: obtaining multidimensional data of the user; performing feature extraction on the multidimensional data to obtain multidimensional features; performing causal enhancement on the multidimensional features and at least one preset intent to obtain causal features; and processing the multidimensional features and causal features through the intent recognition model to obtain a target intent recognition result.

[0163] The storage unit 1420 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 14201 and / or a cache memory unit 14202 , and may further include a read-only memory unit (ROM) 14203 .

[0164] The storage unit 1420 may also include a program / utility 14204 having a set (at least one) of program modules 14205, such program modules 14205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0165] The bus 1430 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0166] Electronic device 1400 may also communicate with one or more external devices 1440 (e.g., a keyboard, pointing device, Bluetooth device, etc.), one or more devices that enable a user to interact with electronic device 1400, and / or any device that enables electronic device 1400 to communicate with one or more other computing devices (e.g., a router, modem, etc.). Such communication may occur via input / output (I / O) interface 1450. Furthermore, electronic device 1400 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network such as the Internet) via network adapter 1460. As shown, network adapter 1460 communicates with other modules of electronic device 1400 via bus 1430. It should be understood that, although not shown, other hardware and / or software modules may be used in conjunction with electronic device 1400, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0167] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0168] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart may be implemented as a computer program product, which includes: a computer program, which implements the above-mentioned intention recognition method when executed by a processor.

[0169] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided. The computer-readable storage medium may be a readable signal medium or a readable storage medium. The computer-readable storage medium stores a program product capable of implementing the above-mentioned method of the present disclosure. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes program code. When the program product is executed on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary embodiments of the present disclosure described in the "Exemplary Methods" section above of this specification.

[0170] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fibers, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0171] In the present disclosure, a computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0172] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination thereof.

[0173] In a specific implementation, the program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a standalone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0174] It should be noted that although several modules or units of the device for action execution are mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0175] Furthermore, although the steps of the method of the present disclosure are described in a particular order in the accompanying drawings, this does not require or imply that the steps must be performed in this particular order, or that all steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0176] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0177] Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art not disclosed herein. The description and examples are to be considered as exemplary only, with the true scope and spirit of the present disclosure being indicated by the appended claims.

Claims

1. A method for identifying intention, characterized in that: include: Obtain multi-dimensional data of users; Performing feature extraction on the multidimensional data to obtain multidimensional features; Performing causal relationship enhancement on the multidimensional features and at least one preset intention to obtain a causal feature; The multidimensional features and the causal features are processed by an intention recognition model to obtain a target intention recognition result.

2. The intention recognition method according to claim 1, characterized in that The step of enhancing the causal relationship between the multidimensional feature and at least one preset intention to obtain a causal feature includes: Based on the multidimensional features and the at least one preset intent, construct a completely undirected graph, where each node in the completely undirected graph represents a feature or an intent, and an edge between two nodes in the completely undirected graph represents a possible dependency relationship between the two nodes; Performing a conditional independence test on each pair of nodes in the completely undirected graph to construct a local undirected graph, wherein a pair of nodes connected by an edge in the local undirected graph are nodes that have a dependency relationship given other nodes; Determine the direction of the edge in the local undirected graph and construct a causal graph; The weight corresponding to each characteristic node in the causal graph is determined to obtain the causal characteristic.

3. The intention recognition method according to claim 2, characterized in that The step of performing a conditional independence test on each pair of nodes in the completely undirected graph to construct a local undirected graph includes: Given other nodes in the completely undirected graph, if a pair of nodes in the completely undirected graph are independent, then the edge between the pair of nodes is deleted, and the pair of nodes includes a feature node and an intention node.

4. The intention recognition method according to claim 3, characterized in that: The step of deleting an edge between a pair of nodes in the completely undirected graph, given other nodes in the completely undirected graph, if the pair of nodes in the completely undirected graph are independent, includes: Under the condition of other nodes in a completely undirected graph, calculating the chi-square value of the one feature node and the one intention node; If the calculated chi-square value is less than or equal to the preset quantity threshold, it is determined that the one feature node and the one intention node are independent.

5. The intention recognition method according to claim 2, characterized in that: Determining the direction of the edge in the local undirected graph and constructing the causal graph includes: A preset inspection method is used to determine the direction of the edge in the undirected graph.

6. The intention recognition method according to claim 2, characterized in that: Determining the weight corresponding to each feature node in the causal graph to obtain the causal feature includes: In a path from a feature node to a path node, if the path is within the second order, determining the weight of the feature node in the path as a first value; If the path is from the intention node to the feature node or the path reaches the third order or above, the weight of the feature node in the path is determined as a second value, wherein the second value is smaller than the first value and the second value is determined by random number generation.

7. The intention recognition method according to claim 1, characterized in that: The intention recognition model includes a first conversion layer and a second conversion layer; The processing of the multidimensional features and the causal features by the intention recognition model to obtain a target intention recognition result includes: Performing semantic recognition on the multidimensional features through a first conversion layer to obtain a semantic vector; Processing the causal features and the semantic vectors through a second conversion layer to obtain probability values ​​of each preset intention; Determine the preset intention with the largest probability value as the target intention recognition result; The intent recognition model is trained by using the user's multi-dimensional historical data and at least one preset intent label.

8. The intention recognition method according to claim 1, characterized in that: After the multidimensional features and the causal features are processed by the intention recognition model to obtain a target intention recognition result, the method includes: Determine a target execution module corresponding to the target intention recognition result, and divert the user to the target execution module.

9. An intention recognition device, characterized in that: include: Data acquisition module, used to obtain multi-dimensional data of users; A feature extraction module, configured to extract features from the multidimensional data to obtain multidimensional features; a feature enhancement module, configured to enhance the causal relationship between the multidimensional feature and at least one preset intention to obtain a causal feature; The intention recognition module is used to process the multidimensional features and the causal features through the intention recognition model to obtain a target intention recognition result.

10. An electronic device, characterized in that: include: processor; as well as a memory for storing executable instructions of the processor; The processor is configured to execute the intention recognition method according to any one of claims 1 to 8 by executing the executable instructions.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the intention recognition method according to any one of claims 1 to 8 is implemented.

12. A computer program product having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the intention recognition method according to any one of claims 1 to 8 is implemented.

Images