User intention recognition method and device, equipment and medium

Abstract

The invention relates to the technical field of data analysis, can be applied to business system platforms of financial science and technology, medical health and the like, and discloses a user intention recognition method, device and equipment and a medium. Generating an intention recognition strategy of each user intention point according to the key intention verbal skill, constructing an intention recognition model, carrying out context semantic analysis on interaction text data of a target user, generating a context perception text, carrying out distributed coding on the key intention verbal skill in the intention recognition strategy, and obtaining intention strategy representation; and determining an attention weight according to the context sensing text and the intention strategy representation, identifying intention key data of the interactive text data, performing intention analysis on the intention key data by using an intention identification model, and generating a target intention of the target user. According to the invention, the accuracy and efficiency of intention recognition of the target user are improved.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a user intent recognition method, apparatus, device, and medium. Background Technology

[0002] In today's insurance industry, with increasingly diverse customer needs, traditional human customer service models face significant challenges. The varying quality of insurance service personnel makes it difficult to fully understand customers' potential needs, especially in non-motor insurance products, where misunderstandings or omissions of customer requirements are common. Because customers' intentions are often not clearly expressed, service personnel frequently rely on their own experience to make judgments. This approach is not only inaccurate but also increases ineffective communication, further delaying problem resolution and severely impacting customer experience. To address these issues, traditional customer service models urgently need to adopt more efficient and accurate methods to identify customer needs, reduce misunderstandings and redundancy in communication, and improve customer service satisfaction.

[0003] In the healthcare field, customer needs are highly complex and involve a great deal of personalization. Relying solely on algorithms may not be enough to fully understand the detailed needs of patients. For example, patients may not express themselves clearly or accurately describe their health conditions during communication, which could lead to misjudgments of their needs by the model, resulting in incorrect service recommendations.

[0004] In the fintech sector, the diversity and complexity of financial products make machine learning algorithms prone to bias when analyzing customer needs. For example, when a customer's expressed needs are not clear enough, the model may make overly one-sided or incorrect judgments, leading to inappropriate product recommendations or service suggestions. This not only reduces customer trust but may also trigger unnecessary financial risks.

[0005] Therefore, current technologies suffer from low accuracy and low efficiency in recognizing the intent of target users. Summary of the Invention

[0006] This invention provides a user intent recognition method, apparatus, device, and medium, the main purpose of which is to solve the problems of low accuracy and low efficiency in recognizing the intent of target users.

[0007] In a first aspect, to achieve the above objective, the present invention provides a user intent recognition method, comprising: Obtain the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements; Construct an intent recognition model based on the user intent points and the intent recognition strategy; The interaction text data of the target user is obtained, and contextual semantic analysis is performed on the interaction text data to generate context-aware text. The key intent statements in the intent recognition strategy are distributedly encoded to obtain the intent strategy representation; Attention weights are determined based on the context-aware text and the intent strategy representation, and the attention weights are used to identify key intent data of the interactive text data. The intent recognition model is used to analyze the key intent data to generate the target intent of the target user.

[0008] In a second aspect, the present invention also provides a user intent recognition device, comprising: The strategy generation module is used to obtain the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements. The model building module is used to build an intent recognition model based on the user intent points and the intent recognition strategy; The semantic analysis module is used to acquire the interactive text data of the target user, perform contextual semantic analysis on the interactive text data, and generate context-aware text. The strategy encoding module is used to perform distributed encoding of key intent statements in the intent recognition strategy to obtain an intent strategy representation. The attention matching module is used to determine attention weights based on the context-aware text and the intent strategy representation, and to use the attention weights to identify key intent data of the interactive text data. The intent analysis module is used to perform intent analysis on the key intent data using the intent recognition model, and generate the target intent of the target user.

[0009] Thirdly, the present invention also provides an electronic device, the electronic device comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the user intent recognition method described above.

[0010] Fourthly, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the user intent recognition method described above.

[0011] This invention acquires the target user's business needs and key intent statements, identifies user intent points based on the business needs, and generates an intent recognition strategy for each user intent point based on the key intent statements. By combining keyword matching and classification with the key intent statements in the business needs, it can effectively identify diverse user intent points. Personalized intent recognition strategies are generated based on user characteristics and intent points, improving the accuracy and targeting of intent recognition. An intent recognition model is constructed based on the user intent points and intent recognition strategies. Through loss function optimization, the model can accurately distinguish multiple complex intents, achieving intelligent recognition of diverse and fine-grained user needs. The invention also acquires the target user's interactive text data, performs contextual semantic analysis on the interactive text data, and generates context-aware text, which not only improves the accuracy of downstream tasks such as intent recognition and sentiment analysis. Furthermore, it enhances the system's ability to grasp user intent in multi-turn dialogues and complex scenarios. It performs distributed encoding of key intent phrases in the intent recognition strategy to obtain intent strategy representations, deeply integrating strategy keywords with logical structures to fully capture inherent semantic dependencies and achieve multi-dimensional, fine-grained expression of strategy content. Attention weights are determined based on the context-aware text and the intent strategy representations, and these weights are used to identify key intent data in the interactive text data. By introducing an attention mechanism and combining context-aware text and intent strategy representations, the accuracy and robustness of intent recognition are effectively improved. By introducing confidence analysis and dynamic verification mechanisms, the intent recognition model performs intent analysis on the key intent data to generate the target user's target intent, thus improving the accuracy and efficiency of target user intent recognition. Attached Figure Description

[0012] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a schematic diagram of an application environment for a user intent recognition method according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a user intent recognition method according to an embodiment of the present invention. Figure 3 This is a flowchart illustrating the process of constructing an intent recognition model in a user intent recognition method according to an embodiment of the present invention. Figure 4 This is a schematic diagram of a user intent recognition device according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of an electronic device that implements a user intent recognition method according to an embodiment of the present invention; Figure 6 This is another structural schematic diagram of an electronic device that implements a user intent recognition method according to an embodiment of the present invention.

[0014] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions of this disclosure, and to fully understand and implement the process of how this disclosure applies technical means to solve technical problems and achieve corresponding technical effects, the technical solutions in the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. The embodiments of this disclosure and the various features within them can be combined with each other without conflict, and the resulting technical solutions are all within the protection scope of this disclosure. All other embodiments obtained by those skilled in the art based on the embodiments of this disclosure without creative effort should fall within the protection scope of this disclosure.

[0016] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this disclosure are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this disclosure described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0017] This application provides a user intent recognition method. The execution subject of this user intent recognition method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the device provided in this application: a server, a terminal, or other electronic devices. In other words, the user intent recognition method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server 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 communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0018] This invention provides a user intent recognition method, which can be applied to applications such as... Figure 1In this application environment, the client communicates with the server via a network. The server can obtain the target user's business needs and key intent statements from the client, identify user intent points based on the business needs, and generate intent recognition strategies for each user intent point based on the key intent statements. By combining keyword matching and classification with the key intent statements in the business needs, the server can effectively identify diverse user intent points. Personalized intent recognition strategies are generated based on user characteristics and intent points, improving the accuracy and targeting of intent recognition. An intent recognition model is constructed based on the user intent points and intent recognition strategies. Through loss function optimization, the model can accurately distinguish multiple complex intents, achieving intelligent recognition of diverse and fine-grained user needs. The server also acquires the target user's interactive text data, performs contextual semantic analysis on the interactive text data, and generates context-aware text. This not only improves the accuracy of downstream tasks such as intent recognition and sentiment analysis but also enhances the system's overall performance. The invention possesses the ability to grasp user intent in multi-turn dialogues and complex scenarios. It performs distributed encoding of key intent phrases in the intent recognition strategy to obtain intent strategy representations. This deeply integrates strategy keywords with logical structures, fully capturing inherent semantic dependencies and achieving multi-dimensional, fine-grained expression of strategy content. Attention weights are determined based on the context-aware text and the intent strategy representations. These attention weights are then used to identify key intent data in the interactive text data. By introducing an attention mechanism and combining context-aware text with intent strategy representations, the accuracy and robustness of intent recognition are effectively improved. Furthermore, by introducing confidence analysis and dynamic verification mechanisms, the intent recognition model analyzes the key intent data to generate the target user's target intent, improving the accuracy and efficiency of target user intent recognition. Finally, the target intent is output and fed back to the user client. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0019] The following explanation of this invention relates to the present invention, which effectively improves the accuracy and robustness of intent recognition by introducing confidence analysis and a dynamic verification mechanism. When the confidence level of the recognition result is high, the intent can be directly confirmed, ensuring response efficiency; when the confidence level is low, a verification process is triggered, and the model is incrementally optimized based on subsequent user interactions, achieving continuous learning and self-iteration. While ensuring recognition accuracy, the system's real-time performance and adaptability are also taken into account, improving the accuracy and efficiency of intent recognition for target users.

[0020] Reference Figure 2The diagram shown is a flowchart illustrating a user intent recognition method according to an embodiment of the present invention. In this embodiment, the user intent recognition method includes: S1. Obtain the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements.

[0021] In this embodiment of the invention, a standard insurance dataset is obtained by denoising the insurance information dataset of the target user. Based on the dataset, feature label analysis is performed to extract the user's identity feature labels. Keyword matching and classification of business needs are performed to identify multiple user intent points. Combining the user's identity feature labels and the key intent dialogue corresponding to each intent point, an intent recognition strategy for each intent point is generated, thereby achieving accurate identification and classification of user intent.

[0022] In specific healthcare scenarios, it can be applied to intelligent triage systems. By denoising and standardizing the health questionnaires and historical medical data filled out by patients online, it extracts identity feature tags such as age and medical history. Combined with the symptoms and needs expressed by users, it performs keyword analysis and intent recognition, such as "headache," "registration," or "traditional Chinese medicine consultation," thereby constructing a precise intent recognition strategy. This helps the system automatically determine the user's intention to seek medical treatment and recommend suitable departments or doctors, improving medical efficiency and experience.

[0023] In specific fintech scenarios, this can be applied to intelligent insurance advisor services. First, the user's historical policy data is cleaned to extract identity feature tags such as income status, family structure, and risk preference. Then, combined with keywords mentioned by the user in the interaction (such as "retirement security", "critical illness", "financial planning"), the insurance intent is identified, and a matching intent recognition strategy is generated to help the system accurately recommend personalized insurance or financial products, thereby improving the intelligence of financial services and customer satisfaction.

[0024] In this embodiment of the invention, the step of identifying user intent points based on the business requirements and generating an intent recognition strategy for each user intent point based on the key intent dialogue includes: Obtain the insurance information dataset of the target user, and perform noise reduction processing on the insurance information dataset to obtain a standard insurance dataset; Based on the standard insurance dataset, feature label analysis is performed on the target user to obtain the target user's identity feature label; The business requirements and key intent statements are matched and categorized using keywords to obtain multiple intent categories; Intent points are generated based on the intent category, and an intent recognition strategy is generated based on the identity feature label and the intent points.

[0025] In detail, business needs refer to the goals or problems that users hope to achieve in a specific business scenario. For example, in an insurance scenario, business needs might be "to purchase critical illness insurance for family members" or "to understand the coverage of existing policies," typically reflecting the user's intent. Key intent phrases refer to keywords or phrases that users frequently use when expressing business needs and have significant intent identification value. These phrases are highly guiding or representative, such as "how to buy," "whether it pays out," "is there any suitable for the elderly," and "want to understand the coverage," and can be used to help identify the specific intent category of the user.

[0026] Acquire a dataset of insurance information for target users, including but not limited to basic customer information (such as name, age, gender, etc.), historical policies, insurance application time, insured amount, coverage details, and insurance application channels. Perform noise reduction on the dataset, removing redundant, missing, inconsistent, or formatted data items, such as duplicate insurance records, invalid fields, or incorrect input information, to ensure data integrity and accuracy. Generate a standardized insurance dataset with a unified structure and clear semantics that can be used for subsequent analysis, providing a reliable data foundation for user profiling and intent recognition.

[0027] Based on standard insurance datasets, feature tag analysis is performed on target users to extract key attribute information in insurance behavior, such as age, gender, occupation, marital status, family composition, types of insurance purchased, protection preferences, and risk tolerance. These attributes are then categorized and coded through a pre-defined tag system to form structured identity feature tags, such as "high-risk preference," "family-oriented customer," and "young and middle-aged individuals primarily purchasing medical insurance." This provides a personalized basis for user profile construction and intent strategy formulation. User segmentation is the first step in user identification, helping to clearly distinguish between high-value users and potential users, as well as loyal users and users prone to churn. By segmenting users, their needs and intentions can be identified more accurately.

[0028] By using a pre-set key intent dialogue database, keyword matching is performed on the business needs expressed by users to identify high-frequency terms or typical expressions related to specific intents in the text, such as "how to buy", "which diseases are covered", "is it suitable for children", "want to cancel the policy", etc. By comparing the semantic alignment of these key dialogues with the user's original needs and classification rules, they are classified into the corresponding intent categories, and finally multiple clear intent points are extracted, such as "insurance consultation", "inquiry about the scope of coverage", "recommendation of children's insurance", "cancellation request", etc., to achieve accurate identification and multi-dimensional classification of user intents.

[0029] By combining the target user's identity characteristics (such as age group, risk preference, and type of insurance purchased) with the identified intent points (such as "policy cancellation request", "inquiry about adding insurance", "children's protection"), a scenario-based model is created for each intent point, matching the corresponding recognition rules and response modes. Through a strategy generation engine, user characteristics and intent types are comprehensively considered to generate personalized intent recognition strategies. For example, users with "high risk preference + asset allocation" are given priority to financial product recommendations, while users with "middle-aged and young adults + policy cancellation" are prompted to retain the option, thereby improving the accuracy of intent recognition and the intelligence of service response.

[0030] By denoising and analyzing user insurance data, a precise user profile is created. Keyword matching and classification, combined with key intent statements in business requirements, effectively identify diverse user intent points. Furthermore, personalized intent recognition strategies are generated based on user characteristics and intent points, improving the accuracy and relevance of intent recognition, enhancing the system's understanding of user needs, and thus enabling more intelligent and precise service recommendations and responses, significantly improving user experience and business efficiency.

[0031] S2. Construct an intent recognition model based on the user intent point and the intent recognition strategy.

[0032] In this embodiment of the invention, by extracting the business type and interaction goal of user intent points, the intent points are classified and hierarchically structured to generate semantically clear intent tags. The key phrases in the intent recognition strategy are transformed into intent semantic vectors, and the structured features are extracted from them. The three together construct a multi-dimensional feature space. On this basis, an initial intent model is trained and optimized iteratively through a preset loss function, ultimately forming a high-precision intent recognition model with semantic understanding and structure awareness capabilities, which is used to improve the system's intelligent recognition and response capabilities to complex user needs.

[0033] In specific healthcare scenarios, it can be applied to intelligent consultation systems. By identifying keywords and intentions expressed by users when describing symptoms (such as "recently coughing" or "needs medication"), and combining these with users' basic health records and medical history tags (such as "elderly" or "chronic disease patient"), user needs can be categorized into specific consultation intentions (such as "respiratory department appointment" or "follow-up visit for medication"). By building an intention recognition model, accurate matching can be achieved, assisting the system in recommending corresponding departments, doctors, or services, thus enabling more efficient online medical services.

[0034] In specific fintech scenarios, it can be applied to intelligent customer service or investment advisory systems. By analyzing users' language expressions during the consultation process (such as "Are there any stable financial products?" or "How do I buy pension insurance?"), combined with feature tags such as insurance records and risk preferences, multiple financial service intent points can be extracted and classified (such as "conservative financial recommendations" or "pension fund allocation needs"). Personalized recommendations or automatic responses can be achieved through intent recognition models, thereby improving the intelligence and response efficiency of services, enhancing customer experience, and increasing business conversion rates.

[0035] Figure 3 This is a flowchart illustrating the process of constructing an intent recognition model in a user intent recognition method according to an embodiment of the present invention.

[0036] In this embodiment of the invention, constructing the intent recognition model based on the user intent point and the intent recognition strategy includes: Extract the business type and interaction goal of the user intent point, classify and hierarchically structure the user intent point according to the business type and interaction goal, and generate an intent tag system; The key intent phrases in the intent recognition strategy are embedded into intent semantic vectors; Extract the structured features from the intent recognition strategy; Construct a multi-dimensional feature space using the intent labeling system, the intent semantic vector, and the structured features; Construct an initial intent model based on the multi-dimensional feature space; The initial intent model is optimized using a preset loss function to obtain an intent recognition model.

[0037] In detail, the business types include: insurance application, renewal, surrender, claims, consultation, underwriting, etc., and the interaction goals include: obtaining product information, submitting application requests, changing contract terms, calculating premiums, etc. The corresponding business type and interaction goal are extracted from each intent point, and these two dimensions jointly characterize the business scenario and specific purpose of the user's current behavior. Based on preset business logic and semantic classification rules, different intent points are classified first-level by business type, and then further refined into second-level categories according to interaction goals, constructing a multi-level, structured intent tag system. For example, "understanding the coverage of children's critical illness insurance" can be categorized as "Consultation > Product Coverage > Children's Critical Illness Insurance," and "Applying for surrender of whole life insurance" can be categorized as "Surrender > Product Surrender > Life Insurance." Semantically identifiable intent tags (such as Consultation_Coverage_Children's Critical Illness, Surrender_Life Insurance) are generated for each categorized intent point to facilitate subsequent feature fusion, model training, and service deployment.

[0038] After generating the intent recognition strategy, the core expressive phrases and keywords used to identify different user intents are extracted, namely key intent phrases (such as "how to buy", "can you compensate", "is it suitable for children"), and the key intent phrases are semantically encoded using pre-trained language models (such as BERT, Word2Vec, ERNIE, etc.), and converted into low-dimensional dense semantic vector representations, preserving contextual meaning and semantic relevance.

[0039] Structured features include: user identity feature tags (such as age group, risk preference), interaction behavior features (such as click path, session rounds, response duration), intent triggering context (such as previous operation, guidance dialogue type), etc. These features have clear field structure and business semantics.

[0040] These structured features are then integrated with the intent semantic vectors generated in the previous step (which reflect the deep semantic information of user speech) and the intent tagging system built based on business type and interaction goal to construct a high-dimensional feature space covering multiple dimensions such as semantic understanding, user behavior, and business context.

[0041] Based on the constructed multi-dimensional feature space, the fused intent labels, semantic vectors, and structured features are used as input. A neural network model (such as a multilayer perceptron, Transformer, or BERT fine-tuning model) is employed to build an initial intent recognition model. By modeling the non-linear relationships between features, the corresponding intent classification result is output. During training, the system uses a preset loss function (such as cross-entropy loss, Focal Loss, etc.) to measure the error between the model's predicted intent and the true intent label, and continuously updates the model parameters through backpropagation to optimize model performance. After multiple rounds of iterative training, a high-precision intent recognition model that can accurately understand user semantics and adapt to various intent scenarios is finally obtained. The loss function calculation formula is shown below:

[0042] in, The total number of intent categories. Indicates the weight of the intent category. Indicates the intention of the first Predicted probability of class Indicates the regulating factor. This represents the total loss value.

[0043] By categorizing and hierarchically structuring user intent points according to their business types and interaction goals, a clear intent tagging system is constructed. Simultaneously, semantic information (semantic vectors embedded in key intent phrases) and structured features (such as user behavior and contextual state) from the intent recognition strategy are integrated to build a multi-dimensional feature space. This enables the model to comprehensively perceive the semantic expression and business context of user intent. Based on this, an initial intent model is built, and through loss function optimization, the model can accurately distinguish multiple complex intents, achieving intelligent recognition of diverse and fine-grained user needs. This significantly improves the system's accuracy and response efficiency in scenarios such as customer service, recommendation, and decision support.

[0044] S3. Obtain the interactive text data of the target user, perform contextual semantic analysis on the interactive text data, and generate context-aware text.

[0045] In this embodiment of the invention, the interactive text is segmented and stop words are removed to extract effective content words. The effective content words and context information are jointly encoded, semantically fused and enhanced using a preset language model to obtain fused text semantics. The interactive text data is labeled based on the fused text semantics to generate context-aware text.

[0046] In specific healthcare scenarios, it can be applied to intelligent understanding of electronic medical records and assistance in doctor consultations. For example, by segmenting and encoding the dialogue text between doctors and patients, and integrating the description of the illness and past medical history information, it can achieve contextual awareness and understanding of the patient's complaints, thereby assisting in the generation of structured medical records, accurately extracting symptoms and diagnostic points, and improving the intelligence level of clinical decision support systems.

[0047] In specific fintech scenarios, it can be applied to intelligent customer service and risk identification. For example, in the interaction between users and financial platforms, the system can identify the user's true intentions through contextual semantic analysis, such as loan inquiries and appeals for abnormal transactions. At the same time, it can enhance semantics by combining the context of user behavior, effectively distinguishing between normal inquiries and potential fraudulent behavior, thereby improving the response efficiency and risk control capabilities of financial services.

[0048] In this embodiment of the invention, the step of performing contextual semantic analysis on the interactive text data to generate context-aware text includes: The interactive text data is segmented to obtain several interactive content words; Remove stop words from the interactive content words to obtain valid content words; Obtain the context information of the interactive text data, and use a preset language model to jointly encode the effective content words and the context information to obtain a context representation vector; The interactive text data and the context representation vector are semantically fused and enhanced to obtain fused text semantics; The interactive text data is annotated based on the fused text semantics to obtain context-aware text.

[0049] In detail, a word segmentation algorithm based on dictionary matching and statistical learning is applied to segment interactive text data, dividing continuous word sequences into basic units with linguistic meaning. A domain-defined dictionary is introduced to identify professional terms and compound words related to specific business scenarios. Based on a set stop word list, prepositions, conjunctions and other function words that contribute little to semantics are removed, and only core words that can express key meanings are retained. The retained words are then labeled with parts of speech and filtered for context to obtain a number of effective content words for subsequent intent recognition or semantic modeling, which can reflect user intent, behavioral tendencies and interactive context semantics.

[0050] By capturing the interactive dialogue history or text window, the contextual information of the current text is extracted, including the user's previous input, the system's response, and related contextual content. The extracted contextual information is semantically associated with the effective content words obtained from the previous word segmentation and used as model input. A pre-set language model (such as BERT or Transformer) is used to jointly encode the effective content words and contextual information. Through a multi-layer attention mechanism, the dependency relationship between words and context is modeled to capture semantic connections and contextual logic, and output a contextual representation vector representing the interactive text in a specific context.

[0051] The current interactive text is encoded into word vector representations through an embedding layer, and then aligned with the context representation vectors obtained through joint encoding for feature alignment and dimension matching. Attention or gating mechanisms are used to dynamically fuse the local semantics of the current text with the global semantic information of the context, strengthening semantic dependence and context constraints. Deep semantic features are further extracted through residual connections or multi-layer semantic enhancement networks to improve the expressive ability of the text in the context and generate fused text semantics containing context-aware features.

[0052] Each word or phrase is accurately classified based on context-aware semantic information. By combining entity features, intent cues, and contextual dependencies in the fused semantics, key positions in the interactive text are labeled word by word, and semantic units such as intent words, slot values, and sentiment tendencies are identified, resulting in labeled context-aware text.

[0053] The formula for calculating the language model is shown below:

[0054] in, This represents context-aware text. Representing a language model, Represents interactive text data, The context-aware text is represented as a 768-dimensional sentence vector. This indicates the fusion of textual semantics.

[0055] By performing contextual semantic analysis on interactive text data to generate context-aware text, the system can fully integrate the user's current input with historical context, enabling semantic understanding to move beyond isolated text and possess stronger contextual relevance and semantic expressiveness. Effective content word extraction enhances semantic focus, while joint encoding and semantic enhancement strengthen context-dependent modeling. The generated context-aware text not only improves the accuracy of downstream tasks such as intent recognition and sentiment analysis but also enhances the system's ability to grasp user intent in multi-turn dialogues and complex scenarios, thereby significantly improving the accuracy and response quality of human-computer interaction.

[0056] S4. Distributed encoding is performed on the key intent statements in the intent recognition strategy to obtain the intent strategy representation.

[0057] In this embodiment of the invention, a complex strategy is decomposed into several strategy units through structured processing, and the strategy keywords and logical structures are extracted. The keywords are embedded with word vectors, and the logical structures are parsed. Based on this, the semantic coupling and logical connection between the key embedded features and the strategy semantic features are captured through the dependency relationship between them. The above information is then uniformly semantically encoded to form a structured and semantically rich intention strategy representation.

[0058] In specific healthcare scenarios, this technology can be applied to intelligent consultation systems. By performing contextual semantic analysis on the symptom descriptions input by patients, key symptom words can be extracted and combined with contextual information such as past medical history to generate context-aware text, thereby improving the accuracy of disease identification. Simultaneously, structuring doctors' pre-set consultation paths or treatment rules into strategy units and semantically encoding them helps build intelligent assisted diagnostic processes, enabling dynamic questioning, risk alerts, and personalized treatment recommendations.

[0059] In specific fintech scenarios, it can be applied to intelligent customer service or intelligent investment advisory systems. By performing semantic fusion and context enhancement on user interaction dialogues, it can accurately identify the user's true intentions in scenarios such as wealth management, loans, and risk consultation. At the same time, it can structure financial service processes and risk control strategies into intent recognition strategies, and transform them into learnable strategy representations through semantic dependency modeling. This enables the system to flexibly respond to changing business requests and achieve personalized recommendations, risk control review assistance, and efficient automatic responses.

[0060] In this embodiment of the invention, the step of distributively encoding the key intent statements in the intent recognition strategy to obtain an intent strategy representation includes: The key intent statements in the intent recognition strategy are divided into several strategy units; Obtain the policy keywords and policy logic structure from the policy unit; Word vector embedding is performed on the keywords of the strategy to obtain key embedding features; The policy logic structure is parsed and encoded to obtain policy semantic features; The dependency relationship between the key embedded features and the policy semantic features is captured to obtain the semantic dependency relationship; Based on the semantic dependencies, the key embedding features and the policy semantic features are semantically encoded to obtain the intent policy representation.

[0061] In detail, the intent recognition strategy is analyzed to identify key elements such as intent target, applicable conditions, triggering statements, and response actions. According to the preset strategy template or semantic structure, each strategy is decomposed into a strategy unit with independent semantic function. Each unit usually contains strategy keywords, semantic tags, and logical relationships. The hierarchical structure and logical dependencies between strategy units are clarified through rule extraction or graph structure modeling, forming a set of clearly structured and coded strategy units.

[0062] Each strategy unit undergoes text parsing and syntactic analysis to identify core words with business implications or decision-oriented meaning, which are then used as strategy keywords, such as action verbs (e.g., "query," "recommendation") and target entities (e.g., "product," "interest rate"). Dependency parsing or rule template matching methods are used to extract the syntactic or semantic structures of the strategy, including conditional statements, triggering logic, and constraint relationships, to construct an abstract representation of the strategy's logical structure. This logical structure typically uses a tree or graph structure to represent the causal relationships, conditional dependencies, or sequential logic between different keywords.

[0063] By calling a pre-trained word vector model (such as Word2Vec, GloVe, or BERT), each policy keyword is mapped to a high-dimensional semantic vector to capture contextual semantics and relationships. All embedding vectors are normalized or aggregated to obtain key embedding features that can be used for subsequent semantic modeling and policy analysis.

[0064] Syntactic analysis and semantic parsing are performed on the policy logic structure to extract key logical elements such as conditions, constraints, and actions. The parsing results are expressed in a structured manner using methods such as rule encoding, graph structure modeling, or sequence modeling. Furthermore, semantic embedding methods are used to vectorize each logical unit, ultimately generating policy semantic features that reflect the policy semantic relationships and execution logic.

[0065] Construct a fusion structure (such as attention mechanism, graph neural network or Transformer encoder), take key embedded features and policy semantic features as joint input, and explore the association patterns between the two in the semantic space through feature alignment, semantic association modeling and other methods. Optimize model parameters through training to strengthen the dependency mapping between key semantic elements and policy logic elements, and finally generate semantic dependency features that can accurately represent their semantic relationship.

[0066] Using semantic dependencies as guiding information, key embedded features and policy semantic features are fused. Feature weighting and interaction enhancement are achieved through attention mechanisms or fusion networks. Semantic encoding modules (such as bidirectional encoders, graph structure encoders, etc.) are used to uniformly model the fused features, extract high-level semantic information, and finally generate an intent-policy representation that can comprehensively reflect user intent and policy logic. The calculation formula is shown below:

[0067] in, Indicates the first Intent strategy representation of class intent, Indicates the first The first key intent phrase for this type of intent, Indicates the first The first class intention A key intention in the dialogue, Indicates the first The key embedding feature corresponding to the first key intent utterance of the class intent, Indicates the first The first class intention Key embedded features corresponding to key intent statements.

[0068] By distributing the encoding of intent recognition strategies, the keywords and logical structure of the strategies can be deeply integrated, fully capturing the inherent semantic dependencies and achieving multi-dimensional and fine-grained expression of strategy content. This not only improves the semantic richness and accuracy of strategy representation, but also enhances the model's ability to understand complex intents and strategy logic, which helps to improve the accuracy and robustness of intent recognition, thereby promoting the optimization of the effects and application expansion of intelligent interaction systems.

[0069] S5. Determine attention weights based on the context-aware text and the intent strategy representation, and use the attention weights to identify key intent data of the interactive text data.

[0070] In this embodiment of the invention, a matching score between context-aware text and intent policy representation is calculated, and attention weights are generated accordingly. The attention weights are then used to perform weighted fusion of context-aware text and intent policy representation to obtain an intent-aware semantic vector. Based on this semantic vector, keywords in the interactive text are identified, and key intent data is extracted.

[0071] In specific healthcare scenarios, this technology can be applied to natural language interactions between patients and health assistants. By extracting the matching relationship between context-aware text and medical intent strategies, it can identify key intent data such as diagnostic requests and symptom descriptions in the user's statements. For example, the system can automatically identify the core health intent based on the patient's input, "I've been feeling dizzy and nauseous for the past few days," and categorize it as "neurological symptoms," thus assisting in intelligent diagnosis and the generation of subsequent treatment suggestions.

[0072] In specific fintech scenarios, it can be applied to the dialogue recognition between users and intelligent customer service or investment advisory assistants. By integrating contextual information and financial intent strategies, it can effectively extract key intent data such as "loan," "interest rate," and "change trend" from phrases like "I want to know about changes in loan interest rates." This allows for the rapid identification of users' financial needs, enabling personalized financial recommendations, risk warnings, or business guidance, thereby improving interaction efficiency and service accuracy.

[0073] In this embodiment of the invention, determining attention weights based on the context-aware text and the intent policy representation, and using the attention weights to identify key intent data of the interactive text data, includes: Obtain the total number of intent categories and the attention parameter matrix, and determine the matching score between the context-aware text and the intent policy representation based on the total number of intent categories and the attention parameter matrix; Generate corresponding attention weights based on the matching scores; The context-aware text and the intent policy representation are fused according to the attention weights to obtain an intent-aware semantic vector; Determine the similarity between the intent-aware semantic vector and each of the interactive text data, and filter out the interactive text data corresponding to the similarity greater than a preset similarity threshold; The selected interactive text data will be used as key intent data.

[0074] In detail, the total number of preset intent categories and the corresponding attention parameter matrix are obtained to measure the similarity and discriminability between different intents. Context-aware text and intent strategy representations are input into the attention parameter matrix, and the matching scores of the two under different intent categories are calculated to reflect semantic relevance. Attention weights are generated based on the distribution of matching scores to highlight semantic information highly relevant to the target intent. The formula for calculating attention weights is as follows:

[0075] in, This represents context-aware text. Represents the attention parameter matrix, Indicates the first Intent strategy representation of class intent, The total number of intent categories. Indicates the first Intent strategy representation of class intent, This represents the attention weight.

[0076] Using the attention weights generated in the previous step, the semantic features in the context-aware text and intent strategy representation are weighted and processed. The two are then organically combined in the semantic space through a weighted fusion operation, thereby enhancing the semantic representation's ability to focus on the user's current intent and ultimately generating an intent-aware semantic vector that can comprehensively reflect the user's intent.

[0077] Guided by intent-aware semantic vectors, semantic relevance scores are scored for each word or phrase in interactive text data. The words most closely related to the current intent are identified as keywords. By comparing the similarity between the semantic vectors and the word vectors in the text, the content that plays a core role in expressing the user's intent is selected, and finally, the key intent data is extracted.

[0078] By introducing an attention mechanism and combining context-aware text with intent strategy representation, the accuracy and robustness of intent recognition are effectively improved. By calculating semantic matching scores and generating attention weights, key semantic information is focused on. Furthermore, by calculating similarity with the interactive text, content highly relevant to the target intent is filtered out, accurately extracting key intent data. This not only improves the understanding of complex intents but also enhances the system's response accuracy in multi-turn interactions.

[0079] S6. Utilize the intent recognition model to perform intent analysis on the key intent data and generate the target intent of the target user.

[0080] In this embodiment of the invention, potential intents are extracted and matched with corresponding intent categories. The average and standard deviation are calculated by combining historical intent confidence data to generate a confidence threshold. If the current intent confidence is lower than the confidence threshold, a preset review mechanism is triggered for manual or automatic verification. The model is then incrementally learned by combining updated interaction data to achieve adaptive optimization. If the current intent confidence is higher than the confidence threshold, the matching result is directly confirmed as the target intent, thereby achieving an efficient and continuously optimized user intent recognition process.

[0081] In specific healthcare scenarios, this technology can be applied to intelligent consultation systems. By analyzing key data such as patient symptom descriptions and medical history, it can extract potential medical intentions (e.g., "seeking diabetes treatment" or "making an appointment with a cardiologist") and determine the confidence level by combining this with historical patient intention data. If the confidence level is low, it is reviewed by a doctor or the system, and the model is optimized using subsequent interaction data to improve the system's ability to respond to diverse health needs.

[0082] In specific fintech scenarios, this technology can be applied to intelligent customer service or investment advisory systems to identify user intent in statements such as "I want to transfer money" or "Check investment returns." When the confidence level of the identified intent is insufficient, the system can trigger a manual review process to ensure the security of financial operations. At the same time, the recognition model is continuously trained through subsequent user interaction data to achieve more accurate and personalized intent recognition, thereby improving customer service efficiency and compliance.

[0083] In this embodiment of the invention, the step of using the intent recognition model to perform intent analysis on the key intent data and generating the target intent of the target user includes: The intent recognition model is used to extract the latent intent from the key intent data; Identify the matching intent categories of the potential intent, and perform confidence analysis on the matching intent categories to obtain the intent confidence; Obtain historical intent confidence data of the intent recognition model, determine the average value of the historical intent confidence data, and determine the standard deviation based on the average value; A confidence threshold is generated based on the mean and the standard deviation; Determine whether the confidence level of the intent is greater than or equal to the confidence threshold; If the confidence level of the intent is less than the confidence level threshold, the matching intent category is detected according to a preset verification mechanism to obtain a verification result; Collect updated interaction data of the target user, and perform incremental learning on the intent recognition model based on the updated interaction data and the review result to obtain an updated recognition model; The updated recognition model is used to perform intent recognition on the key intent data to obtain the target user's target intent. If the confidence level of the intent is greater than or equal to the confidence threshold, then the matching intent category is taken as the target intent of the target user.

[0084] In detail, the intent recognition model is used to perform semantic parsing on key intent data, extracting the implicit potential intent information. By comparing with a preset intent labeling system, the most matching intent category is identified. Based on the probability distribution or similarity score output by the model, the confidence of the matching results is analyzed to quantify the degree of consistency between the intent category and the user's actual intent, thus providing a basis for subsequent judgment on whether further review is needed. The formula for calculating intent confidence is as follows:

[0085] in, Represents interactive text data, Indicates the first Attention weights for intent categories Indicates the category of matching intent. Indicates learnable weights, This indicates that the intent category in the interactive text data is numbered. Class intent confidence.

[0086] By retrieving the historical intent recognition results accumulated in the system, the corresponding intent confidence values ​​are extracted to form a historical intent confidence dataset. Based on this, the average value of the historical intent confidence dataset is calculated to reflect the overall reliability level of past recognition results. The standard deviation is further calculated based on the average value to measure the range of confidence fluctuation, providing a statistical basis for constructing a dynamic confidence threshold. The formula for calculating the confidence threshold is as follows:

[0087] in, Indicates the confidence threshold. This represents the average value. It represents the standard deviation.

[0088] If the confidence level of the intent is less than the confidence level threshold, the matched intent category is detected according to a preset verification mechanism. The verification mechanism may include manual review, expert rule verification or auxiliary model verification, etc. By comparing and analyzing the user's original interaction content with the recognition results, it is determined whether the matched intent is accurate and a verification result is generated to guide the model to adjust or confirm the final intent.

[0089] After the review is completed, continuously collect updated interaction data from the target user, including user responses to system feedback, new questions or instructions, etc. This updated data, along with the review results, is used as labeled samples for incremental learning of the intent recognition model. By fine-tuning the model parameters or efficiently updating with a small number of samples, the model's accuracy in recognizing similar low-confidence intents is gradually improved, resulting in an updated recognition model that better reflects user behavior patterns. This achieves adaptive optimization of the model. The formula for the incremental learning framework is shown below:

[0090] in, This indicates updating the model parameters of the recognition model. The model parameters represent the intent recognition model. Indicates the learning rate. Represents the loss function Regarding parameters gradient, This indicates an update to the interaction data. This indicates the result of the review.

[0091] After the model completes incremental learning and updates, the updated recognition model is reused to analyze the original key intent data. Through optimized semantic understanding and classification capabilities, the potential intent features in the user's expression are extracted and matched with the most suitable intent category, ultimately generating a more accurate target user intent result, thereby improving the overall recognition effect and user interaction experience.

[0092] If the confidence level of the intent is greater than or equal to the confidence threshold, the currently matched intent category is considered to have sufficient credibility. Without further verification, the matched intent category is directly determined as the target user's target intent and used as the basis for subsequent service processes or response decisions, thereby achieving fast and efficient intent recognition and response.

[0093] By introducing confidence analysis and a dynamic verification mechanism, the accuracy and robustness of intent recognition are effectively improved. When the confidence level of the recognition result is high, the intent can be directly confirmed, ensuring response efficiency; when the confidence level is low, a verification process is triggered, and the model is incrementally optimized based on subsequent user interactions, achieving continuous learning and self-iteration. While ensuring recognition accuracy, the system also considers real-time performance and adaptability, making it suitable for diverse and demanding intelligent interaction scenarios.

[0094] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0095] like Figure 4The diagram shown is a functional block diagram of a user intent recognition device provided in an embodiment of the present invention.

[0096] In this embodiment of the disclosure, a user intent recognition device is provided, which corresponds one-to-one with the user intent recognition method described in the above embodiments. For example... Figure 4 As shown, the user intent recognition device 100 can be installed in an electronic device. Depending on its functions, the user intent recognition device 100 includes a policy generation module 101, a model building module 102, a semantic analysis module 103, a policy encoding module 104, an attention matching module 105, and an intent analysis module 106. Detailed descriptions of each functional module are as follows: The strategy generation module 101 is used to acquire the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements. Model building module 102 is used to build an intent recognition model based on the user intent points and the intent recognition strategy; Semantic analysis module 103 is used to acquire the interactive text data of the target user, perform contextual semantic analysis on the interactive text data, and generate context-aware text; The strategy encoding module 104 is used to perform distributed encoding of key intent statements in the intent recognition strategy to obtain an intent strategy representation. Attention matching module 105 is used to determine attention weights based on the context-aware text and the intent strategy representation, and to use the attention weights to identify key intent data of the interactive text data; The intent analysis module 106 is used to perform intent analysis on the key intent data using the intent recognition model to generate the target intent of the target user.

[0097] In one embodiment, the strategy generation module 101 performs the following steps: identifying user intent points based on the business requirements and generating an intent recognition strategy for each user intent point based on the key intent dialogue. Obtain the insurance information dataset of the target user, and perform noise reduction processing on the insurance information dataset to obtain a standard insurance dataset; Based on the standard insurance dataset, feature label analysis is performed on the target user to obtain the target user's identity feature label; The business requirements and key intent statements are matched and categorized using keywords to obtain multiple intent categories; Intent points are generated based on the intent category, and an intent recognition strategy is generated based on the identity feature label and the intent points.

[0098] In one embodiment, the model building module 102, when performing the action of building an intent recognition model based on the user intent point and the intent recognition strategy, includes: Extract the business type and interaction goal of the user intent point, classify and hierarchically structure the user intent point according to the business type and interaction goal, and generate an intent tag system; The key intent phrases in the intent recognition strategy are embedded into intent semantic vectors; Extract the structured features from the intent recognition strategy; Construct a multi-dimensional feature space using the intent labeling system, the intent semantic vector, and the structured features; Construct an initial intent model based on the multi-dimensional feature space; The initial intent model is optimized using a preset loss function to obtain an intent recognition model.

[0099] In one embodiment, the semantic analysis module 103 performs contextual semantic analysis on the interactive text data to generate context-aware text, including: The interactive text data is segmented to obtain several interactive content words; Remove stop words from the interactive content words to obtain valid content words; Obtain the context information of the interactive text data, and use a preset language model to jointly encode the effective content words and the context information to obtain a context representation vector; The interactive text data and the context representation vector are semantically fused and enhanced to obtain fused text semantics; The interactive text data is annotated based on the fused text semantics to obtain context-aware text.

[0100] In one embodiment, the policy encoding module 104 performs distributed encoding of key intent statements in the intent recognition policy to obtain an intent policy representation, including: The key intent statements in the intent recognition strategy are divided into several strategy units; Obtain the policy keywords and policy logic structure from the policy unit; Word vector embedding is performed on the keywords of the strategy to obtain key embedding features; The policy logic structure is parsed and encoded to obtain policy semantic features; The dependency relationship between the key embedded features and the policy semantic features is captured to obtain the semantic dependency relationship; Based on the semantic dependencies, the key embedding features and the policy semantic features are semantically encoded to obtain the intent policy representation.

[0101] In one embodiment, the attention matching module 105 determines attention weights based on the context-aware text and the intent policy representation, and uses the attention weights to identify key intent data of the interactive text data, including: Obtain the total number of intent categories and the attention parameter matrix, and determine the matching score between the context-aware text and the intent policy representation based on the total number of intent categories and the attention parameter matrix; Generate corresponding attention weights based on the matching scores; The context-aware text and the intent policy representation are fused according to the attention weights to obtain an intent-aware semantic vector; Determine the similarity between the intent-aware semantic vector and each of the interactive text data, and filter out the interactive text data corresponding to the similarity greater than a preset similarity threshold; The selected interactive text data will be used as key intent data.

[0102] In one embodiment, the intent analysis module 106 performs intent analysis on the key intent data using the intent recognition model to generate the target user's target intent, including: The intent recognition model is used to extract the latent intent from the key intent data; Identify the matching intent categories of the potential intent, and perform confidence analysis on the matching intent categories to obtain the intent confidence; Obtain historical intent confidence data of the intent recognition model, determine the average value of the historical intent confidence data, and determine the standard deviation based on the average value; A confidence threshold is generated based on the mean and the standard deviation; Determine whether the confidence level of the intent is greater than or equal to the confidence threshold; If the confidence level of the intent is less than the confidence level threshold, the matching intent category is detected according to a preset verification mechanism to obtain a verification result; Collect updated interaction data of the target user, and perform incremental learning on the intent recognition model based on the updated interaction data and the review result to obtain an updated recognition model; The updated recognition model is used to perform intent recognition on the key intent data to obtain the target user's target intent. If the confidence level of the intent is greater than or equal to the confidence threshold, then the matching intent category is taken as the target intent of the target user.

[0103] In this invention, a user intent recognition device is first developed by acquiring the target user's business needs and key intent phrases. Based on the business needs, the device identifies user intent points and generates an intent recognition strategy for each intent point based on the key intent phrases. By combining keyword matching and classification with the key intent phrases in the business needs, the device effectively identifies diverse user intent points. Personalized intent recognition strategies are generated based on user characteristics and intent points, improving the accuracy and targeting of intent recognition. An intent recognition model is constructed based on the user intent points and the intent recognition strategies. Through loss function optimization, the model can accurately distinguish multiple complex intents, achieving intelligent recognition of diverse and fine-grained user needs. Then, the device acquires the target user's interactive text data and performs contextual semantic analysis on the interactive text data to generate context-aware text, further enhancing intent recognition and emotion perception. The system improves the accuracy of downstream tasks such as analysis and enhances its ability to grasp user intent in multi-turn dialogues and complex scenarios. It performs distributed encoding of key intent phrases in the intent recognition strategy to obtain intent strategy representations, deeply integrating strategy keywords with logical structures to fully capture inherent semantic dependencies and achieve multi-dimensional, fine-grained expression of strategy content. Attention weights are determined based on the context-aware text and the intent strategy representations, and these weights are used to identify key intent data in the interactive text data. By introducing an attention mechanism and combining context-aware text and intent strategy representations, the accuracy and robustness of intent recognition are effectively improved. Finally, by introducing confidence analysis and dynamic verification mechanisms, the intent recognition model is used to analyze the key intent data and generate the target user's target intent, improving the accuracy and efficiency of target user intent recognition. Specific limitations of a user intent recognition device can be found in the limitations of a user intent recognition method described above, and will not be repeated here. Each module in the above-described user intent recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or they can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of the above modules.

[0104] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When executed by the processor, the computer program implements a user intent recognition method, a server-side function, or step.

[0105] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 6 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When executed by the processor, the computer program implements a user intent recognition method's client-side functions or steps.

[0106] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: Obtain the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements; Construct an intent recognition model based on the user intent points and the intent recognition strategy; The interaction text data of the target user is obtained, and contextual semantic analysis is performed on the interaction text data to generate context-aware text. The key intent statements in the intent recognition strategy are distributedly encoded to obtain the intent strategy representation; Attention weights are determined based on the context-aware text and the intent strategy representation, and the attention weights are used to identify key intent data of the interactive text data. The intent recognition model is used to analyze the key intent data to generate the target intent of the target user.

[0107] In the several embodiments provided by this invention, it should be understood that the disclosed devices and apparatuses can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0108] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0109] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0110] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0111] In some embodiments of this example, a computer-readable storage medium is provided, on which a computer program is stored, characterized in that the computer program, when executed by a processor, implements the steps of the method described in the above embodiments.

[0112] The readable storage medium of the present invention stores a computer program, which, when executed by a processor of an electronic device, can perform the following: Obtain the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements; Construct an intent recognition model based on the user intent points and the intent recognition strategy; The interaction text data of the target user is obtained, and contextual semantic analysis is performed on the interaction text data to generate context-aware text. The key intent statements in the intent recognition strategy are distributedly encoded to obtain the intent strategy representation; Attention weights are determined based on the context-aware text and the intent strategy representation, and the attention weights are used to identify key intent data of the interactive text data. The intent recognition model is used to analyze the key intent data to generate the target intent of the target user.

[0113] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0114] Computer-readable storage media may also store at least one computer-executable program / instruction, such as computer-readable instructions. Computer-readable storage media include, but are not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Computer-readable storage media may include, for example, read-only memory (ROM), hard disk, flash memory, etc. For example, a non-transitory computer-readable storage medium may be connected to a computing device such as a computer, and then, when the computing device executes the computer-readable instructions stored on the computer-readable storage medium, the various methods described above can be performed.

[0115] In addition, the computer device may include (but is not limited to) a data bus, an input / output (I / O) bus, a display, and input / output devices (e.g., keyboard, mouse, speakers, etc.).

[0116] The processor can communicate with external devices via the I / O bus through wired or wireless networks.

[0117] In one embodiment, the at least one computer-executable instruction may also be compiled into or comprise a software product / computer program product, wherein one or more computer-executable instructions are executed by a processor to perform the steps of the various functions and / or methods in the embodiments described herein.

[0118] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), Synchlink, DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0120] In the embodiments provided in this disclosure, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0121] It should be noted that, in this disclosure, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element limited by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0122] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

[0123] It should be noted that if any software tools or components not belonging to our company appear in the embodiments of this application, they are merely for illustrative purposes and do not represent actual use.

Claims

1. A method for recognizing user intent, characterized in that, The method includes: Obtain the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements; Construct an intent recognition model based on the user intent points and the intent recognition strategy; The interaction text data of the target user is obtained, and contextual semantic analysis is performed on the interaction text data to generate context-aware text. The key intent statements in the intent recognition strategy are distributedly encoded to obtain the intent strategy representation; Attention weights are determined based on the context-aware text and the intent strategy representation, and the attention weights are used to identify key intent data of the interactive text data. The intent recognition model is used to analyze the key intent data to generate the target intent of the target user.

2. The user intent recognition method as described in claim 1, characterized in that, The step of identifying user intent points based on the business requirements and generating an intent recognition strategy for each user intent point based on the key intent dialogue includes: Obtain the insurance information dataset of the target user, and perform noise reduction processing on the insurance information dataset to obtain a standard insurance dataset; Based on the standard insurance dataset, feature label analysis is performed on the target user to obtain the target user's identity feature label; The business requirements and key intent statements are matched and categorized using keywords to obtain multiple intent categories; Intent points are generated based on the intent category, and an intent recognition strategy is generated based on the identity feature label and the intent points.

3. The user intent recognition method as described in claim 1, characterized in that, The step of constructing an intent recognition model based on the user intent point and the intent recognition strategy includes: Extract the business type and interaction goal of the user intent point, classify and hierarchically structure the user intent point according to the business type and interaction goal, and generate an intent tag system; The key intent phrases in the intent recognition strategy are embedded into intent semantic vectors; Extract the structured features from the intent recognition strategy; Construct a multi-dimensional feature space using the intent labeling system, the intent semantic vector, and the structured features; Construct an initial intent model based on the multi-dimensional feature space; The initial intent model is optimized using a preset loss function to obtain an intent recognition model.

4. The user intent recognition method as described in claim 1, characterized in that, The step of performing contextual semantic analysis on the interactive text data to generate context-aware text includes: The interactive text data is segmented to obtain several interactive content words; Remove stop words from the interactive content words to obtain valid content words; Obtain the context information of the interactive text data, and use a preset language model to jointly encode the effective content words and the context information to obtain a context representation vector; The interactive text data and the context representation vector are semantically fused and enhanced to obtain fused text semantics; The interactive text data is annotated based on the fused text semantics to obtain context-aware text.

5. The user intent recognition method as described in claim 1, characterized in that, The distributed encoding of key intent statements in the intent recognition strategy to obtain an intent strategy representation includes: The key intent statements in the intent recognition strategy are divided into several strategy units; Obtain the policy keywords and policy logic structure from the policy unit; Word vector embedding is performed on the keywords of the strategy to obtain key embedding features; The policy logic structure is parsed and encoded to obtain policy semantic features; The dependency relationship between the key embedded features and the policy semantic features is captured to obtain the semantic dependency relationship; Based on the semantic dependencies, the key embedding features and the policy semantic features are semantically encoded to obtain the intent policy representation.

6. The user intent recognition method as described in claim 1, characterized in that, The step of determining attention weights based on the context-aware text and the intent strategy representation, and using the attention weights to identify key intent data of the interactive text data, includes: Obtain the total number of intent categories and the attention parameter matrix, and determine the matching score between the context-aware text and the intent policy representation based on the total number of intent categories and the attention parameter matrix; Generate corresponding attention weights based on the matching scores; The context-aware text and the intent policy representation are fused according to the attention weights to obtain an intent-aware semantic vector; Determine the similarity between the intent-aware semantic vector and each of the interactive text data, and filter out the interactive text data corresponding to the similarity greater than a preset similarity threshold; The selected interactive text data will be used as key intent data.

7. The user intent recognition method as described in claim 1, characterized in that, The step of using the intent recognition model to perform intent analysis on the key intent data and generating the target intent of the target user includes: The intent recognition model is used to extract the latent intent from the key intent data; Identify the matching intent categories of the potential intent, and perform confidence analysis on the matching intent categories to obtain the intent confidence; Obtain historical intent confidence data of the intent recognition model, determine the average value of the historical intent confidence data, and determine the standard deviation based on the average value; A confidence threshold is generated based on the mean and the standard deviation; Determine whether the confidence level of the intent is greater than or equal to the confidence threshold; If the confidence level of the intent is less than the confidence level threshold, the matching intent category is detected according to a preset verification mechanism to obtain a verification result; Collect updated interaction data of the target user, and perform incremental learning on the intent recognition model based on the updated interaction data and the review result to obtain an updated recognition model; The updated recognition model is used to perform intent recognition on the key intent data to obtain the target user's target intent. If the confidence level of the intent is greater than or equal to the confidence threshold, then the matching intent category is taken as the target intent of the target user.

8. A user intent recognition device, characterized in that, The device includes: The strategy generation module is used to obtain the target user's business needs and key intent statements, identify user intent points based on the business needs, and generate an intent recognition strategy for each user intent point based on the key intent statements. The model building module is used to build an intent recognition model based on the user intent points and the intent recognition strategy; The semantic analysis module is used to acquire the interactive text data of the target user, perform contextual semantic analysis on the interactive text data, and generate context-aware text. The strategy encoding module is used to perform distributed encoding of key intent statements in the intent recognition strategy to obtain an intent strategy representation. The attention matching module is used to determine attention weights based on the context-aware text and the intent strategy representation, and to use the attention weights to identify key intent data of the interactive text data. The intent analysis module is used to perform intent analysis on the key intent data using the intent recognition model, and generate the target intent of the target user.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the user intent recognition method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the user intent recognition method as described in any one of claims 1 to 7.

Images