Business data intelligent sorting method and device, equipment and medium

Abstract

The present application relates to the technical field of data analysis, and discloses a business data intelligent sorting method, device, equipment and medium. The method comprises the following steps: fusing structured features and unstructured features of business data corresponding to a plurality of preset businesses into target business features, and converting the target business features into business feature vectors; analyzing viewing probabilities of the plurality of preset businesses according to a first network branch, analyzing business demand intention labels by using a second network branch, analyzing business handling probabilities by using a third network branch, and calculating comprehensive priority feature values of the business data in the plurality of preset businesses; sorting the business data of the plurality of preset businesses according to the comprehensive priority feature values, generating an initial business sequence, re-sorting the business data based on real-time feedback data returned by the initial business sequence, and generating a target business sequence. The present application can be applied to financial technology, medical health and other business system platforms, and can improve the accuracy of intelligent sorting of business data.

Description

Technical Field

[0001] This invention relates to the field of data analysis technology, and in particular to a method, apparatus, device, and medium for intelligent sorting of business data. Background Technology

[0002] As my country's population continues to age, the demand for elderly care services is becoming increasingly diversified and personalized. Currently, various elderly care service platforms typically push elderly care service information to users in list format, and the rationality of the push order directly affects the user's information acquisition efficiency and service conversion rate.

[0003] Most existing elderly care service recommendation and ranking schemes rely on a single-dimensional ranking strategy. For example, some schemes rank service information solely based on click-through rate (CTR) estimates, prioritizing services with higher historical CTR. However, this approach ignores users' actual business needs and can easily lead to high click-through rates but low conversion rates.

[0004] In addition, most existing solutions are offline static sorting models. Once the sorting results are generated, they are fixed and applied to the push scenario. There is a lack of online correction mechanism for real-time feedback data. When user behavior patterns change or new services are launched, the model cannot adjust the sorting strategy in time, causing the push effect to gradually decline over time.

[0005] For example, in the healthcare context, as residents' awareness of health management continues to increase, healthcare platforms have accumulated a large amount of structured data (such as laboratory indicators and medication records) and unstructured data (such as medical record texts and chief complaint descriptions). Existing recommendation rankings often rely on single indicators such as click-through rate or historical visit frequency, ignoring patients' true medical intentions, and lack effective integration of structured and unstructured features, resulting in a low degree of matching between recommended content and actual health needs.

[0006] For example, in fintech scenarios, fintech platforms aggregate both structured user data (such as asset size and risk rating) and unstructured data (such as consultation records and demand descriptions). Recommendation ranking directly impacts product conversion rates and user experience. Most existing ranking models are offline and static, lacking online correction mechanisms based on real-time transaction feedback. When market conditions or user preferences change, ranking strategies cannot adapt in a timely manner, leading to a continuous decline in recommendation accuracy.

[0007] Therefore, how to implement an intelligent sorting method for business data to improve the accuracy of business data sorting and thus enhance the precision of business push is a technical problem that urgently needs to be solved. Summary of the Invention

[0008] This invention provides a method, apparatus, device, and medium for intelligent sorting of business data, the main purpose of which is to solve the problem of low accuracy in sorting business data.

[0009] Firstly, to achieve the above objectives, the present invention provides an intelligent sorting method for business data, comprising: Obtain business data corresponding to multiple preset services, extract the structured features of the business data in each preset service, and extract the unstructured features of the business data in each preset service; Using a pre-defined cross-modal attention mechanism, the structured and unstructured features of the business data in each pre-defined service are fused to obtain the target business features of the business data in each pre-defined service, and each target business feature is converted into a business feature vector of the business data in each pre-defined service. A multi-branch neural network is obtained, which includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis. Obtain the historical viewing probability of target personnel for business data in each preset business within a preset historical time period. Utilize the first network branch to perform viewing probability analysis on business data in each preset business based on the historical time period and the historical viewing probability, and predict the corresponding viewing probability of business data in each preset business within the target time period. The second network branch is used to perform demand intent analysis on the business feature vector to generate business demand intent labels for the business data in each preset business. The third network branch is used to perform business processing probability analysis on the business data in each preset business based on the business feature vector, and to predict the business processing probability of the business data in each preset business within the target time period. Calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent, and the business processing probability; All the business data are sorted in descending order according to the comprehensive priority feature value to generate an initial business sequence; The initial service sequence is pushed to the user terminal, and real-time feedback data returned by the user terminal is received. Based on the real-time feedback data, the comprehensive priority feature value of the service data in each preset service is corrected. All the business data are reordered based on the corrected comprehensive priority feature value to generate the target business sequence.

[0010] Secondly, the present invention also provides a business data intelligent sorting device, comprising: The feature extraction module is used to acquire business data corresponding to multiple preset business, extract the structured features of the business data in each preset business, and extract the unstructured features of the business data in each preset business. The feature fusion module is used to perform feature fusion on the structured and unstructured features of the business data in each preset service using a preset cross-modal attention mechanism, to obtain the target business features of the business data in each preset service, and to convert each target business feature into a business feature vector of the business data in each preset service. The branch acquisition module is used to acquire a multi-branch neural network, which includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis. The probability analysis module is used to obtain the historical viewing probability of the target personnel for each preset business data within a preset historical time period. The first network branch is used to perform a viewing probability analysis on the business data of each preset business based on the historical time period and the historical viewing probability, and to predict the viewing probability of the business data of each preset business within the target time period. The demand intent analysis module is used to perform demand intent analysis on the business feature vector using the second network branch, and generate business demand intent tags for business data in each preset business. The probability analysis module is used to perform business processing probability analysis on the business data in each preset business based on the business feature vector using the third network branch, and to predict the business processing probability of the business data in each preset business within the target time period. The priority calculation module is used to calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent, and the business processing probability. The initial business sorting module sorts all the business data in descending order according to the comprehensive priority feature value to generate an initial business sequence. The priority correction module is used to push the initial service sequence to the user terminal and receive real-time feedback data returned by the user terminal, and correct the comprehensive priority feature value of the service data in each preset service based on the real-time feedback data. The business target sorting module is used to reorder all the business data according to the corrected comprehensive priority feature value to generate a target business sequence.

[0011] 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 intelligent sorting method for business data described above.

[0012] 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 above-described intelligent sorting method for business data.

[0013] In this embodiment of the invention, structured and unstructured features are integrated and a cross-modal attention mechanism is utilized to enable the computer to comprehensively understand business attributes, thereby improving the discriminative power and robustness of feature representation. Secondly, multi-branch neural networks are used to model viewing probabilities, demand intentions, and processing probabilities in parallel, achieving multi-objective joint prediction of user "interest-intent-conversion," which is better able to capture the intrinsic relationships between complex behaviors compared to single-task models. Based on these three factors, a comprehensive priority feature value is calculated and sorted in descending order, transforming the multi-dimensional evaluation results into an efficient business sequence and reducing the computational complexity of online decision-making. Furthermore, by receiving real-time feedback data to correct priorities and reorder them, a closed-loop online learning mechanism is formed, which can quickly adapt to changes in user preferences and shifts in business popularity, continuously optimizing recommendation accuracy. This enables the distribution of business data for different preset businesses with high matching degrees under limited display resources, significantly improving the accuracy of business data sorting. Attached Figure Description

[0014] 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.

[0015] Figure 1 This is a schematic diagram of an application environment for the intelligent sorting method for business data in one embodiment of the present invention. Figure 2 A flowchart illustrating a business data intelligent sorting method according to an embodiment of the present invention; Figure 3 A functional block diagram of a business data intelligent sorting device provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of the structure of an electronic device for implementing an intelligent sorting method for business data according to an embodiment of the present invention; Figure 5 This is another structural schematic diagram of an electronic device for implementing an intelligent sorting method for business data, as provided in an embodiment of the present invention.

[0016] 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

[0017] It should be noted that in the technical solutions disclosed in this invention, the acquisition of user information (personal image data (e.g., facial videos or pictures, facial feature videos or pictures, etc.) and personal privacy information (e.g., name, ID number, occupation, address, etc.)) is all completed with the user's knowledge and consent, and the acquisition of the relevant user information is legal and compliant.

[0018] 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.

[0019] It should be noted that the terms "first," "second," etc., used in this disclosure and the accompanying drawings 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.

[0020] This application provides a method for intelligently sorting business data. The executing entity of this 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 equipment. In other words, the intelligent sorting 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.

[0021] The intelligent sorting method for business data of this invention can be applied to, for example... Figure 1 In this application environment, the client communicates with the server via a network. 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 now be described in detail through specific embodiments.

[0022] Reference Figure 2 The diagram shown is a flowchart illustrating a business data intelligent sorting method according to an embodiment of the present invention. In this embodiment, the business data intelligent sorting method includes: S1. Obtain business data corresponding to multiple preset services, extract the structured features of the business data in each preset service, and extract the unstructured features of the business data in each preset service.

[0023] In this embodiment of the invention, multiple preset services include elderly care services, and the service data includes structured and unstructured data corresponding to the respective services. The structured data refers to standardized service data in different preset services that has a fixed format, regular fields, and can be directly stored in tables / databases, such as service prices, service duration, elderly person's age, number of residents, fee standards, and service type codes. The unstructured data refers to service data in different preset services that lacks a unified fixed format and cannot be directly categorized by field, such as service evaluation texts, elderly person's verbal needs, nursing logs, on-site recordings, service photos, customer complaint texts, promotional texts, and follow-up voice messages.

[0024] The structured features include, but are not limited to, the service scale and service type contained in the business data of each preset business, while the unstructured features include, but are not limited to, user needs, sentiment tendencies, and need types.

[0025] In this embodiment of the invention, the extraction of structured features from the business data in each preset business includes: Extract the structured data of the business data in each preset business, and parse the field type of each field in the structured data; Based on the field type, each field of the structured data is divided into numeric fields, categorical fields, and date fields; Calculate the numerical statistical characteristics of the numerical field; One-hot encoding is performed on the category field and the date field to obtain discrete encoding features; The numerical statistical feature vector and the discrete coding feature are concatenated to generate the structured features of the business data in each preset business.

[0026] In this embodiment of the invention, each field value of the structured data in the business data of each preset business is scanned column by column. The data type of this column is determined by regular expression matching, value range distribution analysis and sample sampling. For example, if the values ​​of a column are all numbers with decimal points, it is determined to be a numeric type; if they are all limited text labels such as "town / rural", it is determined to be a category type; if the time format such as year-month-day or year and month appears, it is determined to be a date type.

[0027] In detail, three independent field lists are constructed based on data type. After traversing all fields, each field is placed into its corresponding list. Specifically, these can be divided into numeric fields, categorical fields, and date fields. For each numeric field, a set of statistics is calculated to characterize its overall distribution, including the average level, maximum value, minimum value, median value, and degree of fluctuation of all values ​​in this field. For each categorical and date field, all possible values ​​of this field are first counted. Then, an independent position is created for each possible value. If a record's field value happens to be equal to a certain value, it is marked as valid in the corresponding position, and all other positions are marked as invalid. The same applies to date fields; the time granularity, such as year-month-day or year-month, is broken down into independent categories and then encoded in the same way to convert it into a numeric vector with only two states: valid and invalid.

[0028] Furthermore, by connecting them end to end in a fixed order, they are combined into a complete long vector. This long vector contains both statistical summary information of numerical fields and discrete identity information of categorical and date fields, ultimately becoming a unified structured feature representation of business data records in each preset business, which can be directly input into subsequent machine learning models.

[0029] In this embodiment of the invention, the extraction of unstructured features from the business data in each preset business includes: Extract unstructured data from the business data in each preset business, identify the data type of the unstructured data, and match the corresponding feature extraction model from the preset feature extraction model library according to the data type; If the data type of the unstructured data is text, then the feature extraction model is a natural language processing model. The unstructured data is encoded with word vectors according to the natural language processing model to obtain text semantic features. If the data type of the unstructured data is an image, then the feature extraction model is a convolutional neural network model. The pixel values ​​of the unstructured data are standardized according to the convolutional neural network model to obtain image features. The text semantic features and the image features are concatenated to obtain the unstructured features of the business data.

[0030] In this embodiment of the invention, the type of unstructured data (text or image) is determined by the file extension, file header information, or content sampling. For example, if a file ends with a text suffix and its content can be read as text, it is determined to be text; if it is a common image format and the pixel matrix can be decoded, it is determined to be image. After determination, a search and matching is performed in a pre-maintained model library: for text types, a natural language processing model is invoked; for image types, a convolutional neural network model is invoked.

[0031] Specifically, the text content of the entire business data in the preset business is input into the natural language processing model. The model will first break each word into smaller language units, and then use a multi-layer attention mechanism to understand the contextual meaning of each word in the whole sentence. Finally, the entire text is compressed into a dense vector of fixed length.

[0032] Specifically, pixel value standardization involves scaling the color value of each pixel in an image from its original range to a uniform standard range, eliminating numerical fluctuations caused by differences in shooting light and equipment. After standardization, the image is fed into a convolutional neural network. The network slides and scans the image through multiple layers of convolutional kernels, extracting visual information at different levels, from simple edges and textures to complex object contours, and finally outputs a fixed-length vector at the deep layer of the network.

[0033] Furthermore, the text semantic vector and the image feature vector are concatenated end to end in a predetermined order to form a longer vector, so that when the subsequent model processes business data in the preset business, it can understand both the text content and the image content. The two complement each other to form a complete unstructured data feature representation.

[0034] For example, in a healthcare scenario, taking an online consultation platform as an example, structured data includes fields such as the patient's age, gender, previous departments visited, laboratory test values ​​(e.g., blood glucose, blood pressure), and medication records. These fields are mapped to structured feature vectors through feature encoding. Numerical fields are directly concatenated after normalization, while categorical fields are concatenated after embedding mapping. Unstructured data includes patient complaints, medical records, and doctor-patient dialogue records. A pre-trained language model is used to semantically encode these texts, extracting sentence-level or paragraph-level unstructured feature vectors. For instance, for a complaint containing recurrent headaches and dizziness over the past week and a history of hypertension, the model extracts semantic features related to headaches and hypertension. For a structured record containing a blood glucose level of 8.2 mmol / L and a consultation with an endocrinologist, numerical features and department category features are extracted. After these two types of features are extracted, they proceed to the subsequent cross-modal fusion stage.

[0035] For example, in a fintech scenario, taking a bank wealth management product recommendation platform as an example, structured data includes fields such as user asset size, risk tolerance level, types of products purchased historically, and holding period. Feature engineering converts these fields into structured feature vectors, where asset size is normalized after logarithmic transformation, and risk level is mapped to a dense vector after ordered encoding. Unstructured data includes user consultation records, descriptions of wealth management needs, and open-ended answers to risk preference questionnaires. A pre-trained language model is used to semantically encode these texts and extract unstructured feature vectors. For example, for a consultation record expressing a desire for principal protection and stable returns while rejecting high risk, the model extracts semantic features indicating a preference for principal protection and a low-risk inclination. For a structured record containing assets of 500,000 yuan and a conservative risk level, corresponding numerical and categorical features are extracted. After both types of feature extraction are completed, they are fed into a cross-modal attention mechanism for fusion.

[0036] In this embodiment of the invention, by acquiring the structured and unstructured features of the business data in each preset business, the completeness of data representation and the depth of model understanding can be significantly improved. Structured features facilitate efficient statistical analysis and relational modeling by computers, while unstructured features, through technologies such as natural language processing and image recognition, mine the implicit semantics in text or images, effectively supporting subsequent tasks such as intelligent sorting of business data.

[0037] S2. Using a preset cross-modal attention mechanism, feature fusion is performed on the structured and unstructured features of the business data in each preset service to obtain the target business features of the business data in each preset service, and each target business feature is converted into a business feature vector of the business data in each preset service.

[0038] In this embodiment of the invention, the step of using a preset cross-modal attention mechanism to fuse the structured and unstructured features of the business data in each preset service to obtain the target business features of the business data in each preset service includes: Obtain the structured and unstructured features of business data in the same preset business, and map the structured and unstructured features to a feature space of the same dimension to obtain aligned structured and unstructured features; The aligned structured features are used as the query vector, and the aligned unstructured features are used as the key vector and value vector. The dot product similarity between the query vector and the key vector is calculated using a preset cross-modal attention mechanism, and the dot product similarity is normalized to obtain the attention weight distribution. The attention weight distribution and the value vector are weighted and summed to obtain the target business features of the business data in each preset business.

[0039] In this embodiment of the invention, structured and unstructured features from different sources are transformed into the same semantic space to ensure comparability in subsequent attention calculations. Specifically, each type of feature data is processed through its own independent embedding layer or linear transformation layer. These transformation layers consist of trained weight parameters that can convert the original feature data into vector representations. Structured features typically come from tabular data or numerical records, while unstructured features may come from text, images, or other complex data formats, which originally have completely different data structures and dimensional meanings. Through mapping operations, the two types of features are transformed into vectors of the same length and the same semantic reference system. This alignment process ensures that they can effectively interact and compare within the same feature space, laying the foundation for subsequent attention weight calculations.

[0040] Specifically, the feature vector of one modality is selected as the query vector, which is responsible for actively searching and matching relevant information, equivalent to the questioner role in the attention mechanism. The feature vector of the other modality is used as both the key vector and the value vector. The key vector is responsible for providing the feature's identity information for matching calculation with the query vector, while the value vector stores the actual content information that needs to be extracted and utilized. This allows the same feature to play different functions under different roles, thereby realizing feature interaction and semantic association mining between modalities.

[0041] In detail, after the query vector and key vector are generated, their matching degree needs to be calculated to obtain the attention weights. This matching degree is calculated by summing the products of corresponding elements of the two vectors, reflecting the extent to which the query vector is associated with a certain key vector. The larger the sum of the products, the higher the matching degree. After calculating all the dot product similarity scores, these scores need to be normalized so that the sum of all scores equals a fixed value, thereby transforming the original scores into a reasonable weight distribution.

[0042] After obtaining the attention weight distribution, the attention weight of each feature is multiplied by its corresponding value vector, and then all products are summed to obtain the fused target business feature. This process achieves dynamic feature integration; the influence of each original feature on the final result is determined by its corresponding attention weight. High-importance features are amplified, while low-importance features are suppressed. Through this weighted summation operation, feature information from different modalities is organically fused together. The generated new feature retains the unique information of each modality while reflecting the interrelationships between modalities. The final target business feature can more comprehensively and accurately reflect the overall status of business data in each preset business.

[0043] In this embodiment of the invention, a pre-defined cross-modal attention mechanism is used to adaptively weight and fuse the structured and unstructured features of business data in each pre-defined business. This dynamically captures the complementary and correlated information between the two types of features, effectively suppressing redundancy or noise interference. The resulting target business features are more compact and semantically richer, and are then converted into a unified business feature vector, thereby enhancing the overall accuracy of intelligent sorting of business data.

[0044] S3. Obtain a multi-branch neural network, wherein the multi-branch neural network includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis.

[0045] In this embodiment of the invention, the multi-branch neural network adopts a design concept that combines parameter sharing and task separation. The bottom layer is a common feature extraction layer responsible for extracting common features from the original business data. Then, these common features are fed into three independent network branches. Each branch has its own dedicated fully connected layer and activation function, which is specifically designed for deep modeling of a single task. This structure allows the network to learn three different targets at the same time: viewing probability, demand intent, and processing probability, without interference between them due to differences between tasks.

[0046] Specifically, the first network branch is used for probability analysis. After receiving the business feature vector output by the common feature extraction layer, this branch performs nonlinear transformation through a multi-layer fully connected network, gradually mapping the features to a high-dimensional space related to the user's viewing behavior. Finally, through an output node, the result is compressed to between zero and one, obtaining the probability value of the user viewing business data in any preset business.

[0047] The second network branch is used for demand intent analysis. This branch also takes business feature vectors as input, but its internal structure is specially designed. It first performs non-linear feature transformation through a fully connected layer, then focuses on the core semantics through a dimensionality reduction layer, and finally scores the probability of each label in the preset intent label set through the output layer, selecting the label with the highest score as the intent result.

[0048] The third network branch is used for business processing probability analysis. After feature input, this branch first evaluates and filters the importance of features, retaining the key features that have the most impact on processing behavior. Then, it generates a processing behavior representation through nonlinear mapping. Finally, it combines external influencing factors for fusion and correction, outputting a processing probability that integrates internal features and the external environment.

[0049] In this embodiment of the invention, by constructing a multi-branch neural network that includes three parallel branches—probability analysis, demand intent analysis, and business processing probability analysis—it is possible to jointly model the feature vectors of business data in the same preset business from different semantic dimensions, thereby improving the accuracy of intelligent sorting of business data.

[0050] S4. Obtain the historical viewing probability of the target personnel for each preset business data within a preset historical time period. Utilize the first network branch to perform viewing probability analysis on the business data in each preset business based on the historical time period and the historical viewing probability, and predict the corresponding viewing probability of the business data in each preset business within the target time period.

[0051] In this embodiment of the invention, the step of using the first network branch to perform viewing probability analysis on the service data in each preset service based on the historical time period and the historical viewing probability, and predicting the viewing probability of the service data in each preset service within the target time period, includes: The historical time period of business data in each preset business is discretized into multiple equally spaced historical time windows, and a historical probability time series corresponding to the historical time window is generated according to the historical viewing probability. The hidden layer of the first network branch is used to extract time-series features from the historical probability time series, generating a time-series feature vector of business data in each preset service; The time-series feature vector is processed by a fully connected layer to output a probability value between 0 and 1 as the initial viewing probability of business data in each preset service. Obtain the time feature code of the target time period, and perform weighted fusion of the time feature code and the preliminary viewing probability to generate the viewing probability of business data in each preset business within the target time period.

[0052] In this embodiment of the invention, within each preset service, the overall continuous service data is divided into time periods based on historical statistical time periods, using uniform time intervals. This breaks the original continuous, unbounded time pattern, forming several independent historical time windows with uniform duration and sequential arrangement, thus achieving discretized reconstruction of the continuous time dimension. After the time window division is completed, based on the real historical viewing probability data corresponding to the service data in each preset service, each historical time window is precisely matched one-to-one with the service viewing probability within that window period. The scattered time dimension information and probability data information are systematically integrated, and the data is arranged in chronological order. Finally, a historical probability time series that can completely reflect the dynamic change of the viewing probability of service data in each preset service over time is formed, providing a standardized and structured time series data foundation for subsequent extraction and analysis of time series features.

[0053] The first network branch, the hidden layer, possesses the capability for feature mining and semantic extraction of time-series data. It can accurately capture the temporal correlation characteristics inherent in time-series data. After the preprocessed historical probability time series is input into the hidden layer, the hidden layer automatically mines various effective information hidden in the time series through multi-layer nonlinear transformation and feature filtering mechanisms. This includes key time-series information such as the correlation patterns of probability changes between different time windows, fluctuation trends, steady-state characteristics, and abnormal change characteristics. At the same time, it filters out meaningless redundant interference information in the data, integrates, compresses, and normalizes the extracted core time-series features, and transforms the scattered time-series feature information into time-series feature vectors with unified dimensions, concise features, and representational capabilities, realizing the transformation and upgrading from raw time-series data to high-dimensional feature information.

[0054] This invention employs feature fitting and probability normalization mapping techniques using fully connected layers in a neural network. The core function of the fully connected layer is to globally integrate and fit the temporal feature vectors extracted from the front end. After the temporal feature vectors are input into the fully connected structure, the network performs comprehensive correlation calculations on the feature information of all dimensions in the vector, establishing the intrinsic connections between different feature dimensions, and integrating the representational value of all temporal features to complete deep feature fusion and secondary filtering. After feature fitting, relying on the network's built-in probability normalization activation mechanism, the fused global features are uniformly mapped to a standard probability value range of zero to one. The magnitude of this range can intuitively correspond to the probability of business viewing behavior occurring. Finally, the output is a result that can preliminarily represent the probability of viewing business data in each preset business, obtaining a preliminary viewing probability without considering the target time characteristics, thus completing the initial transformation from feature vectors to probability results.

[0055] Furthermore, for the target time period to be predicted, the inherent time attribute characteristics of this time period are mined, including time period attributes, time cycle characteristics, and seasonal correlation characteristics. Unstructured time attribute information is transformed into structured time feature codes that can be recognized by machines and participate in calculations through standardized coding methods, accurately representing the unique time characteristics of the target time period. Subsequently, a weighted fusion mechanism is introduced, combining the correlation between business data viewing behavior and time attributes for each preset business, and assigning appropriate weight proportions to the initial viewing probability and time feature codes. This retains the core historical probability patterns obtained from the previous time series feature analysis while incorporating the unique time impact characteristics of the target time period. Finally, the optimized result generates business data viewing probability results for each preset business that fits the target time scenario and combines historical patterns with time adaptability.

[0056] In this embodiment of the invention, by utilizing the first network branch, combining the actual viewing probability of each piece of business data by the target personnel within a historical time period, and introducing time series information for modeling, the time dependence and periodicity of user behavior can be effectively captured, thereby dynamically predicting the viewing probability of each piece of business data within a future target time period.

[0057] S5. Utilize the second network branch to perform demand intent analysis on the business feature vector, and generate business demand intent tags for the business data in each preset business.

[0058] In this embodiment of the invention, the step of using the second network branch to perform demand intent analysis on the service feature vector and generating service demand intent tags for each preset service data includes: The first fully connected layer of the second network branch is used to perform a nonlinear transformation on the service feature vector to generate a first hidden feature vector. The second fully connected layer of the second network branch is used to perform feature dimensionality reduction on the first hidden feature vector to generate the second hidden feature vector; The second hidden feature vector is processed by a fully connected method based on a preset set of intent tags to obtain the probability distribution of each of the business feature vectors belonging to each intent tag in the set of intent tags; The intent label with the highest probability value is selected from the probability distribution as the business requirement intent label for the business data in each preset service.

[0059] In this embodiment of the invention, the first fully connected layer of the second network branch performs weighted summation and bias shift on the input business feature vector, and then introduces nonlinear capabilities through an activation function, such as combining scattered fragments of elderly care service demand into implicit semantic directions such as health management preferences. The second fully connected layer maps the first hidden feature vector to a lower-dimensional space. The purpose of dimensionality reduction is twofold: first, to remove redundant information and retain only the features most critical to intent judgment; second, to reduce the computational complexity of subsequent classification, while also achieving a certain regularization effect to prevent overfitting.

[0060] The dimensionality-reduced feature vectors are fed into an output layer. Each node in this layer corresponds to a label in a preset intent label set. Through fully connected and soft normalization operations, each node outputs a value between zero and one. The sum of the values ​​of all nodes is one, forming a complete probability distribution.

[0061] Specifically, from the probability distribution obtained in the previous step, directly find the label with the largest value and use it as the demand intent label for the business data in each preset business.

[0062] In this embodiment of the invention, the second network branch is used to perform in-depth demand intent analysis on the business feature vector, which can automatically map the business data in each preset business to the corresponding business demand intent label, avoiding the subjectivity and inefficiency of manual rule labeling, and improving the accuracy and fineness of label generation.

[0063] S6. Using the third network branch, perform business processing probability analysis on the business data in each preset business based on the business feature vector, and predict the business processing probability of the business data in each preset business within the target time period.

[0064] In this embodiment of the invention, the step of using the third network branch to perform business processing probability analysis on the business data in each preset business based on the business feature vector, and predicting the business processing probability of the business data in each preset business within a target time period, includes: Perform feature importance analysis on the business feature vectors to obtain the feature importance value of each business feature vector; The business feature vectors are filtered based on the feature importance values ​​to obtain target feature vectors; The third network branch is used to perform a nonlinear mapping on the target feature vector to generate a processing behavior representation vector; The processing behavior representation vector is processed by a fully connected layer to obtain a value between 0 and 1 as the initial processing probability. Obtain the vector of external influencing factors within the target time period, and fuse and correct the vector of external influencing factors with the preliminary processing probability to generate the business processing probability corresponding to the business data in each preset business within the target time period.

[0065] In this embodiment of the invention, the network branch first evaluates the importance of each dimension of the input business feature vector, determining which features have a greater impact on the probability of business processing and which are noise or redundancy. A common implementation method is to use attention mechanisms or gradient-based attribution methods to allow the network to automatically learn the weight contribution of each feature, thereby obtaining a set of scores reflecting the importance of each feature. The core purpose of this step is to ensure that subsequent processing focuses on the key features that truly influence user behavior, avoiding interference from irrelevant information.

[0066] Based on the scores obtained in the previous step, a filtering threshold is set, or the top-ranked features are directly selected. Unimportant feature dimensions are removed, and only high-importance features are retained to form a new target feature vector. This process is equivalent to refining and filtering the original features, which reduces the subsequent computational load and allows the network to make predictions in a cleaner feature space, thus improving the reliability of the results.

[0067] The filtered target feature vectors are fed into the hidden layer of the third network branch. Through multi-layer fully connected layers and activation functions, a non-linear transformation is performed to map the features to a high-dimensional space specifically used to express the tendency of handling behavior. The handling behavior representation vector is fully connected through an output node, and then compressed to between zero and one by an activation function to obtain a preliminary handling probability value.

[0068] Specifically, external influencing factors within the target time period, such as policy changes, seasonal factors, and holidays, are acquired and encoded into an external influencing factor vector. This vector is then fused and corrected with the initial processing probability. The fusion method typically involves concatenating the two and then passing them through a fully connected layer, or using multiplication, weighting, or other methods to adjust the initial probability based on the external factors. The final output is a business processing probability that integrates both internal characteristics and the external environment, making the prediction results more closely reflect real-world scenarios.

[0069] In this embodiment of the invention, the third network branch is used to perform business processing probability analysis on the business feature vector. This model can model the deep nonlinear relationship between user historical interactions and business attributes, thereby accurately predicting the processing conversion probability of business data in each preset business within the target time period, significantly improving the conversion prediction accuracy and service efficiency of business data in each preset business.

[0070] S7. Calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent, and the business processing probability.

[0071] In this embodiment of the invention, the viewing probability, business demand intent, and business processing probability have different value ranges and semantic scales. Directly adding them together can lead to one item dominating the result due to its excessively large value. Therefore, it is necessary to map the three to the same interval scale. This is usually achieved by using min-max normalization or soft normalization methods to make each item a comparable relative score, ensuring that the contribution of each dimension is fair during subsequent fusion.

[0072] Specifically, business needs and intentions are transformed into calculable intention weights. Intention itself is a category label, not a numerical value; therefore, a mapping relationship between intention and priority needs to be established first. This is done by statistically analyzing historical data on the actual processing rate and user satisfaction corresponding to different intentions, assigning a weight coefficient to each intention label. For example, health management intentions have a higher weight, while information consultation intentions have a lower weight. In this way, intention transforms from a textual label into a numerical value reflecting its contribution to priority.

[0073] The three aligned values ​​are weighted and summed according to a preset ratio. That is, each value is multiplied by its own weight coefficient and then summed to obtain the comprehensive priority feature value. The weight coefficient can be set by human experience or by training through historical data. The purpose is to ensure that the final result takes into account the three dimensions of whether the user will see it, whether there is a real need, and whether it is easy to process. The absence of any one of these dimensions will not make the priority inflated.

[0074] The comprehensive priority feature value of the business data in each preset business is arranged from high to low, or divided into several levels of high, medium and low according to the threshold, to form the final recommendation ranking result. It can directly determine which business data in the preset business to show to users based on the priority, realizing a complete closed loop from multi-dimensional analysis to executable decision-making.

[0075] For example, in a healthcare scenario, taking an online consultation platform as an example, for each consultation service (such as an appointment with a cardiology specialist), the probability of viewing the network output is predicted based on click-through rate (e.g., 0.85), the intent matching score of the network output business demand is analyzed (e.g., 0.92, indicating a high correlation with the user's chief complaint of "recurring chest tightness"), and the probability of completing the network output business is predicted based on conversion (e.g., 0.78, indicating the likelihood of the user completing the online consultation after clicking). A comprehensive priority feature value is used to calculate a weighted score of 0.869. The viewing probability has the highest weight because users need to see the service first; the intent matching weight is second, ensuring the recommendation is relevant to the actual condition; the completion probability has the lowest weight but cannot be ignored, avoiding recommending services that users view but will not actually seek medical attention for. The consultation services are arranged in descending order of their comprehensive scores to generate an initial recommendation sequence.

[0076] For example, in a fintech scenario, taking bank wealth management product recommendations as an example, for each wealth management product (such as a certain stable fund), the network output viewing probability is predicted through click-through rate (e.g., 0.72), the network output business demand intent matching score is analyzed through intent analysis (e.g., 0.88, indicating a high degree of consistency with the user's wealth management demand of "preserving capital and seeking stability"), and the network output business transaction probability is predicted through conversion (e.g., 0.65, indicating the likelihood of the user completing a purchase after viewing). A comprehensive score of 0.7425 is calculated based on the combined priority feature values. The weight allocation logic is the same as above: viewing probability ensures exposure efficiency, intent matching ensures that the product aligns with the user's risk preferences, and the transaction probability filters out invalid recommendations that are only viewed but not purchased. Each wealth management product is arranged in descending order of its comprehensive score to generate an initial recommendation sequence.

[0077] In this embodiment of the invention, by integrating three heterogeneous outputs—viewing probability, business demand intent tag, and business processing probability—the comprehensive priority feature value of business data in each preset business is calculated. This enables multi-objective joint optimization of the business recommendation order and improves the accuracy of intelligent sorting of business data.

[0078] S8. Sort all the business data in descending order according to the comprehensive priority feature value to generate an initial business sequence.

[0079] In this embodiment of the invention, the comprehensive priority feature value of the business data corresponding to each preset business is bound to its corresponding business identifier to form a business list containing complete priority information. This step provides an operable data foundation for subsequent sorting operations, ensuring that each business does not lose its identity information during the sorting process and can still accurately trace back to the original business after sorting is complete.

[0080] Specifically, the overall priority feature value of all services in the list is compared pairwise, and services with higher feature values ​​are ranked first, while services with lower feature values ​​are ranked last. The comparison process will traverse the entire list until all services are arranged in descending order. This step is the core of the whole process and directly determines the order in which services are displayed. The higher the priority of a service, the earlier it appears and the more likely it is to be seen by the user first.

[0081] In detail, after sorting, the sorted business items are extracted one by one to form a business queue with a clear order. This is the initial business sequence. The business item at the beginning of the sequence has the highest overall priority, and the business item at the end of the sequence has the lowest priority.

[0082] In this embodiment of the invention, the business data in all preset services are sorted in descending order according to the comprehensive priority feature value. This can transform the multi-dimensional evaluation results into a clear and executable one-dimensional business sequence, realize priority-driven automated decision-making, and eliminate the need to repeatedly call complex models for online reordering, thus significantly reducing system computational overhead and response latency.

[0083] S9. Push the initial service sequence to the user terminal and receive real-time feedback data returned by the user terminal. Based on the real-time feedback data, correct the comprehensive priority feature value of the service data in each preset service.

[0084] In this embodiment of the invention, a sorted list of services is sent to the user's mobile phone or other terminal device in the form of a recommendation stream. The user sees information cards of service data in each preset service in descending order of priority. Each interaction of the user after seeing a recommended service is recorded, including whether they clicked to view a service, how long they stayed on a service, whether they directly processed the service, and whether they chose to skip or close it. This behavioral data is transmitted back to the server in real time through a data tracking and reporting mechanism, constituting a feedback signal reflecting the user's true preferences.

[0085] Specifically, user behavior is compared with previous predictions. If a service predicted as high priority is skipped by the user, it indicates that the service's true value has been overestimated and its priority needs to be lowered. Conversely, if a low-priority service is frequently clicked and processed by the user, it indicates that its value has been underestimated and its priority needs to be increased. The correction process typically employs online learning or incremental updates, allowing priority feature values ​​to continuously approach true preferences based on user feedback, forming a continuously optimizing closed loop that makes subsequent recommendations increasingly accurate.

[0086] For example, in a healthcare scenario, taking an online consultation platform as an example, the initial business sequence is a comprehensive sorted list of consultation services (such as specialist appointments, department recommendations, and health checkup packages), which is pushed to patient users via an app. Real-time feedback data generated by patients during use includes: whether they clicked on a doctor's appointment card (click feedback), whether they completed an online consultation after clicking (conversion feedback), whether they gave a positive review or actively repurchased services from the same department after the consultation (satisfaction feedback), and the dwell time after skipping a recommendation (negative feedback). This feedback data is transmitted back in real time, and the comprehensive priority characteristic value of each consultation service is adjusted online based on the feedback signals. For example, if a cardiology specialist is initially ranked high, but users frequently exit without completing the consultation, their comprehensive priority characteristic value is lowered; while a general practitioner, although initially ranked low, has a high conversion rate and positive review rate after clicking, their comprehensive priority characteristic value is increased, and the corrected sequence is re-pushed to subsequent users.

[0087] For example, in a fintech scenario, taking a bank wealth management product recommendation platform as an example, the initial business sequence is a list of wealth management products (such as fixed-term wealth management, funds, insurance, etc.) after comprehensive sorting, which is pushed to users via mobile banking. Real-time feedback data generated by users during use includes: whether they clicked on the details page of a certain product (click feedback), whether they completed the risk assessment and entered the purchase process after clicking (conversion feedback), whether they redeemed the product before maturity after purchase (negative feedback), and the duration and depth of their stay on the product page (interest feedback). The above data is collected in real time and transmitted back, and the comprehensive priority characteristic value of each wealth management product is adjusted accordingly. For example, a certain stable fund may have a high initial ranking, but if a large number of users redeem it early after clicking, it indicates that the actual risk of the product does not match the user's expectations, so its comprehensive priority characteristic value is reduced; while an insurance product may have a low initial ranking, but if users complete the purchase after clicking and hold it until now, its comprehensive priority characteristic value is increased, and the corrected sequence is pushed again.

[0088] In this embodiment of the invention, an initial business sequence is pushed to the user terminal and real-time feedback data is received. Based on this, the comprehensive priority feature value of the business data in each preset business is dynamically corrected, forming a closed-loop online learning mechanism. Compared with a static ranking model, this mechanism significantly improves the system's adaptability to changes in user interests and shifts in business popularity, effectively alleviating information cocoon and recommendation lag problems. This enables the business data recommendation system in each preset business to achieve self-optimization and evolution, improving the reliability of intelligent ranking of business data.

[0089] S10. Reorder all the business data according to the corrected comprehensive priority feature value to generate the target business sequence.

[0090] In this embodiment of the invention, the comprehensive priority feature value of each preset service after updating the service data is obtained from the feedback correction process in the previous step. These values ​​have incorporated the user's actual interaction behavior and are closer to the user's true preferences than the predicted values ​​at the time of initial sorting. The same comparison sorting method as the initial sorting is adopted, but this time the comparison object is the corrected priority feature value instead of the original value. Since the priority of some services has been adjusted upwards or downwards due to user feedback, services that were originally ranked higher may be overtaken by those that were ranked lower, and services that were originally ranked lower may jump significantly due to frequent clicks. After the re-sorting is completed, the new service queue is used as the final recommendation result, replacing the initial sequence previously pushed to the user.

[0091] The position of each service in the target service sequence has been calibrated twice based on user feedback. The highest priority service is placed at the front to ensure that when the user opens the terminal again, they will see the recommended service data from the preset services that best match their current needs after real-time optimization. The entire process forms a complete closed loop from prediction to feedback to correction.

[0092] In this embodiment of the invention, the business data in all preset services are reordered according to the corrected comprehensive priority feature value. This can quickly transform the user dynamic preferences contained in the real-time feedback into adjustments to the order of business display. The reordering mechanism enables the system to immediately correct unreasonable sorting caused by data sparsity or model bias in the initial sequence, thereby improving the timeliness and accuracy of the recommendation results.

[0093] For example, in a healthcare scenario, a high-end senior living community receives approximately 2,000 business data entries monthly from its official website, telephone, and partner channels. The sales team needs to efficiently screen and follow up with high-potential clients. Structured data includes: Client A, age 72, annual income 1.2 million, 3 inquiries; unstructured data includes inquiries such as "My father is in the early stages of Alzheimer's disease and needs professional dementia care. Does your community have a dedicated area and medical support? Can we visit soon?"; and social media comments such as "I saw on the forum that your community environment is very good. I would like to learn about specific dementia care solutions."

[0094] The multimodal fusion engine identifies high urgency (dementia care needs), strong intent (actively requesting a visit), and emotional state (concerned but rational). It then uses a multi-branch neural network to perform hybrid target ranking and optimization, predicting a 98% viewing probability, understanding intent (dementia care, urgency 4.5 / 5), and predicting an 89% business processing probability. Based on the current dynamic weights (urgent intent has a higher weight among recently signed customers), the comprehensive priority score is calculated to be 92.5. The business data is then intelligently ranked according to the comprehensive priority score.

[0095] 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.

[0096] like Figure 3 The diagram shown is a functional block diagram of a business data intelligent sorting device provided in an embodiment of the present invention.

[0097] In this embodiment of the disclosure, a business data intelligent sorting device is provided, which corresponds one-to-one with the business data intelligent sorting method of the above embodiments. For example... Figure 3 As shown, this intelligent business data sorting device 100 can be installed in an electronic device. According to its functions, the intelligent business data sorting device 100 includes a feature extraction module 101, a feature fusion module 102, a branch acquisition module 103, a viewing probability analysis module 104, a demand intent analysis module 105, a processing probability analysis module 106, a priority calculation module 107, an initial business sorting module 108, a priority correction module 109, and a business target sorting module 110. Detailed descriptions of each functional module are as follows: The feature extraction module 101 is used to acquire business data corresponding to multiple preset services, extract the structured features of the business data in each preset service, and extract the unstructured features of the business data in each preset service. The feature fusion module 102 is used to perform feature fusion on the structured and unstructured features of the business data in each preset service using a preset cross-modal attention mechanism, to obtain the target business features of the business data in each preset service, and to convert each target business feature into a business feature vector of the business data in each preset service. The branch acquisition module 103 is used to acquire a multi-branch neural network, which includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis. The probability analysis module 104 is used to obtain the historical viewing probability of the target personnel for each preset business data within a preset historical time period. The first network branch is used to perform viewing probability analysis on the business data of each preset business based on the historical time period and the historical viewing probability, and to predict the viewing probability of the business data of each preset business within the target time period. The demand intent analysis module 105 is used to perform demand intent analysis on the business feature vector using the second network branch, and generate business demand intent tags for business data in each preset business. The probability analysis module 106 is used to perform business processing probability analysis on the business data in each preset business based on the business feature vector using the third network branch, and to predict the business processing probability of the business data in each preset business within the target time period. Priority calculation module 107 is used to calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent and the business processing probability; The initial business sorting module 108 sorts all the business data in descending order according to the comprehensive priority feature value to generate an initial business sequence. The priority correction module 109 is used to push the initial service sequence to the user terminal and receive real-time feedback data returned by the user terminal, and correct the comprehensive priority feature value of the service data in each preset service based on the real-time feedback data. The business target sorting module 110 is used to reorder all the business data according to the corrected comprehensive priority feature value to generate a target business sequence.

[0098] In one embodiment, when the feature extraction module 101 extracts the structured features of the business data in each preset business, it is used to: Extract the structured data of the business data in each preset business, and parse the field type of each field in the structured data; Based on the field type, each field of the structured data is divided into numeric fields, categorical fields, and date fields; Calculate the numerical statistical characteristics of the numerical field; One-hot encoding is performed on the category field and the date field to obtain discrete encoding features; The numerical statistical feature vector and the discrete coding feature are concatenated to generate the structured features of the business data in each preset business.

[0099] In one embodiment, when the feature extraction module 101 extracts unstructured features from the business data in each preset business, it is used to: Extract unstructured data from the business data in each preset business, identify the data type of the unstructured data, and match the corresponding feature extraction model from the preset feature extraction model library according to the data type; If the data type of the unstructured data is text, then the feature extraction model is a natural language processing model. The unstructured data is encoded with word vectors according to the natural language processing model to obtain text semantic features. If the data type of the unstructured data is an image, then the feature extraction model is a convolutional neural network model. The pixel values ​​of the unstructured data are standardized according to the convolutional neural network model to obtain image features. The text semantic features and the image features are concatenated to obtain the unstructured features of the business data.

[0100] In one embodiment, when the feature fusion module 102 performs feature fusion on the structured and unstructured features of the business data in each preset service using a preset cross-modal attention mechanism to obtain the target business features of the business data in each preset service, it is used to: Obtain the structured and unstructured features of business data in the same preset business, and map the structured and unstructured features to a feature space of the same dimension to obtain aligned structured and unstructured features; The aligned structured features are used as the query vector, and the aligned unstructured features are used as the key vector and value vector. The dot product similarity between the query vector and the key vector is calculated using a preset cross-modal attention mechanism, and the dot product similarity is normalized to obtain the attention weight distribution. The attention weight distribution and the value vector are weighted and summed to obtain the target business features of the business data in each preset business.

[0101] In one embodiment, when the viewing probability analysis module 104 performs viewing probability analysis on the service data in each preset service based on the historical time period and the historical viewing probability using the first network branch, and predicts the viewing probability of the service data in each preset service within the target time period, it is used to: The historical time period of business data in each preset business is discretized into multiple equally spaced historical time windows, and a historical probability time series corresponding to the historical time window is generated according to the historical viewing probability. The hidden layer of the first network branch is used to extract time-series features from the historical probability time series, generating a time-series feature vector of business data in each preset service; The time-series feature vector is processed by a fully connected layer to output a probability value between 0 and 1 as the initial viewing probability of business data in each preset service. Obtain the time feature code of the target time period, and perform weighted fusion of the time feature code and the preliminary viewing probability to generate the viewing probability of business data in each preset business within the target time period.

[0102] In one embodiment, when the demand intent analysis module 105 performs demand intent analysis on the service feature vector using the second network branch to generate service demand intent tags for service data in each preset service, it is used to: The first fully connected layer of the second network branch is used to perform a nonlinear transformation on the service feature vector to generate a first hidden feature vector. The second fully connected layer of the second network branch is used to perform feature dimensionality reduction on the first hidden feature vector to generate the second hidden feature vector; The second hidden feature vector is processed by a fully connected method based on a preset set of intent tags to obtain the probability distribution of each of the business feature vectors belonging to each intent tag in the set of intent tags; The intent label with the highest probability value is selected from the probability distribution as the business requirement intent label for the business data in each preset service.

[0103] In one embodiment, when the probability analysis module 106 performs probability analysis on the business data in each preset business based on the business feature vector using the third network branch, and predicts the probability of business processing for each preset business data within a target time period, it is used to: Perform feature importance analysis on the business feature vectors to obtain the feature importance value of each business feature vector; The business feature vectors are filtered based on the feature importance values ​​to obtain target feature vectors; The third network branch is used to perform a nonlinear mapping on the target feature vector to generate a processing behavior representation vector; The processing behavior representation vector is processed by a fully connected layer to obtain a value between 0 and 1 as the initial processing probability. Obtain the vector of external influencing factors within the target time period, and fuse and correct the vector of external influencing factors with the preliminary processing probability to generate the business processing probability corresponding to the business data in each preset business within the target time period.

[0104] In this invention, the specific limitations of the intelligent sorting device for business data can be found in the above-described limitations of the intelligent sorting method for business data, and will not be repeated here. Each module in the aforementioned intelligent sorting device for business data can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0105] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As 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 the computer program is executed by the processor, it implements the functions or steps of the intelligent sorting method for business data on the server side.

[0106] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 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 the computer program is executed by the processor, it implements the functions or steps of the intelligent sorting method for business data on the client side.

[0107] 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 business data corresponding to multiple preset services, extract the structured features of the business data in each preset service, and extract the unstructured features of the business data in each preset service; Using a pre-defined cross-modal attention mechanism, the structured and unstructured features of the business data in each pre-defined service are fused to obtain the target business features of the business data in each pre-defined service, and each target business feature is converted into a business feature vector of the business data in each pre-defined service. A multi-branch neural network is obtained, which includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis. Obtain the historical viewing probability of target personnel for business data in each preset business within a preset historical time period. Utilize the first network branch to perform viewing probability analysis on business data in each preset business based on the historical time period and the historical viewing probability, and predict the corresponding viewing probability of business data in each preset business within the target time period. The second network branch is used to perform demand intent analysis on the business feature vector to generate business demand intent labels for the business data in each preset business. The third network branch is used to perform business processing probability analysis on the business data in each preset business based on the business feature vector, and to predict the business processing probability of the business data in each preset business within the target time period. Calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent, and the business processing probability; All the business data are sorted in descending order according to the comprehensive priority feature value to generate an initial business sequence; The initial service sequence is pushed to the user terminal, and real-time feedback data returned by the user terminal is received. Based on the real-time feedback data, the comprehensive priority feature value of the service data in each preset service is corrected. All the business data are reordered based on the corrected comprehensive priority feature value to generate the target business sequence.

[0108] 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.

[0109] 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.

[0110] 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.

[0111] In some embodiments of this example, a computer-readable storage medium is provided, on which a computer program is stored, which, 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: Acquire business data corresponding to multiple preset services, extract the structured features of the business data in each preset service, and extract the unstructured features of the business data in each preset service; Using a pre-defined cross-modal attention mechanism, the structured and unstructured features of the business data in each pre-defined service are fused to obtain the target business features of the business data in each pre-defined service, and each target business feature is converted into a business feature vector of the business data in each pre-defined service. A multi-branch neural network is obtained, which includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis. Obtain the historical viewing probability of target personnel for business data in each preset business within a preset historical time period. Utilize the first network branch to perform viewing probability analysis on business data in each preset business based on the historical time period and the historical viewing probability, and predict the corresponding viewing probability of business data in each preset business within the target time period. The second network branch is used to perform demand intent analysis on the business feature vector to generate business demand intent labels for the business data in each preset business. The third network branch is used to perform business processing probability analysis on the business data in each preset business based on the business feature vector, and to predict the business processing probability of the business data in each preset business within the target time period. Calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent, and the business processing probability; All the business data are sorted in descending order according to the comprehensive priority feature value to generate an initial business sequence; The initial service sequence is pushed to the user terminal, and real-time feedback data returned by the user terminal is received. Based on the real-time feedback data, the comprehensive priority feature value of the service data in each preset service is corrected. All the business data are reordered based on the corrected comprehensive priority feature value to generate the target business sequence.

[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] 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.

[0117] Those skilled in the art will understand that all or part of the processes in 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, and when executed, it can include the processes of the embodiments of the above methods. Furthermore, 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.

[0118] 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.

[0119] 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. 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.

[0120] 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.

[0121] It should be noted that any AI models, software tools, or components not belonging to this company appearing in the embodiments of this application are merely illustrative examples and do not represent actual use. All user personal information involved in the embodiments of this application has been authorized (with the knowledge and consent) by the relevant parties or has been fully authorized by all parties, and the executing entity may obtain it through various legal and compliant means. The collection, storage, use, processing, transmission, provision, and disclosure of the information, data, and signals involved all comply with relevant laws and regulations and do not violate public order and good morals.

Claims

1. A method for intelligent sorting of business data, characterized in that, The method includes: Obtain business data corresponding to multiple preset services, extract the structured features of the business data in each preset service, and extract the unstructured features of the business data in each preset service; Using a pre-defined cross-modal attention mechanism, the structured and unstructured features of the business data in each pre-defined service are fused to obtain the target business features of the business data in each pre-defined service, and each target business feature is converted into a business feature vector of the business data in each pre-defined service. A multi-branch neural network is obtained, which includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis. Obtain the historical viewing probability of target personnel for business data in each preset business within a preset historical time period. Utilize the first network branch to perform viewing probability analysis on business data in each preset business based on the historical time period and the historical viewing probability, and predict the corresponding viewing probability of business data in each preset business within the target time period. The second network branch is used to perform demand intent analysis on the business feature vector to generate business demand intent labels for the business data in each preset business. The third network branch is used to perform business processing probability analysis on the business data in each preset business based on the business feature vector, and to predict the business processing probability of the business data in each preset business within the target time period. Calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent, and the business processing probability; All the business data are sorted in descending order according to the comprehensive priority feature value to generate an initial business sequence; The initial service sequence is pushed to the user terminal, and real-time feedback data returned by the user terminal is received. Based on the real-time feedback data, the comprehensive priority feature value of the service data in each preset service is corrected. All the business data are reordered based on the corrected comprehensive priority feature value to generate the target business sequence.

2. The intelligent sorting method for business data as described in claim 1, characterized in that, The extraction of structured features from the business data in each preset business includes: Extract the structured data of the business data in each preset business, and parse the field type of each field in the structured data; Based on the field type, each field of the structured data is divided into numeric fields, categorical fields, and date fields; Calculate the numerical statistical characteristics of the numerical field; One-hot encoding is performed on the category field and the date field to obtain discrete encoding features; The numerical statistical feature vector and the discrete coding feature are concatenated to generate the structured features of the business data in each preset business.

3. The intelligent sorting method for business data as described in claim 1, characterized in that, The extraction of unstructured features from the business data in each preset business includes: Extract unstructured data from the business data in each preset business, identify the data type of the unstructured data, and match the corresponding feature extraction model from the preset feature extraction model library according to the data type; If the data type of the unstructured data is text, then the feature extraction model is a natural language processing model. The unstructured data is encoded with word vectors according to the natural language processing model to obtain text semantic features. If the data type of the unstructured data is an image, then the feature extraction model is a convolutional neural network model. The pixel values ​​of the unstructured data are standardized according to the convolutional neural network model to obtain image features. The text semantic features and the image features are concatenated to obtain the unstructured features of the business data.

4. The intelligent sorting method for business data as described in claim 1, characterized in that, The method utilizes a pre-defined cross-modal attention mechanism to fuse the structured and unstructured features of the business data in each pre-defined service, obtaining the target business features of the business data in each pre-defined service, including: Obtain the structured and unstructured features of business data in the same preset business, and map the structured and unstructured features to a feature space of the same dimension to obtain aligned structured and unstructured features; The aligned structured features are used as the query vector, and the aligned unstructured features are used as the key vector and value vector. The dot product similarity between the query vector and the key vector is calculated using a preset cross-modal attention mechanism, and the dot product similarity is normalized to obtain the attention weight distribution. The attention weight distribution and the value vector are weighted and summed to obtain the target business features of the business data in each preset business.

5. The intelligent sorting method for business data as described in claim 1, characterized in that, The step of using the first network branch to perform viewing probability analysis on the service data in each preset service based on the historical time period and the historical viewing probability, and predicting the viewing probability of the service data in each preset service within the target time period, includes: The historical time period of business data in each preset business is discretized into multiple equally spaced historical time windows, and a historical probability time series corresponding to the historical time window is generated according to the historical viewing probability. The hidden layer of the first network branch is used to extract time-series features from the historical probability time series, generating a time-series feature vector of business data in each preset service; The time-series feature vector is processed by a fully connected layer to output a probability value between 0 and 1 as the initial viewing probability of business data in each preset service. Obtain the time feature code of the target time period, and perform weighted fusion of the time feature code and the preliminary viewing probability to generate the viewing probability of business data in each preset business within the target time period.

6. The intelligent sorting method for business data as described in claim 1, characterized in that, The step of using the second network branch to perform demand intent analysis on the business feature vector and generating business demand intent tags for each preset business data includes: The first fully connected layer of the second network branch is used to perform a nonlinear transformation on the service feature vector to generate a first hidden feature vector. The second fully connected layer of the second network branch is used to perform feature dimensionality reduction on the first hidden feature vector to generate the second hidden feature vector; The second hidden feature vector is processed by a fully connected method based on a preset set of intent tags to obtain the probability distribution of each of the business feature vectors belonging to each intent tag in the set of intent tags; The intent label with the highest probability value is selected from the probability distribution as the business requirement intent label for the business data in each preset service.

7. The intelligent sorting method for business data as described in claim 1, characterized in that, The step of using the third network branch to perform business processing probability analysis on the business data in each preset business based on the business feature vector, and predicting the business processing probability of the business data in each preset business within a target time period, includes: Perform feature importance analysis on the business feature vectors to obtain the feature importance value of each business feature vector; The business feature vectors are filtered based on the feature importance values ​​to obtain target feature vectors; The third network branch is used to perform a nonlinear mapping on the target feature vector to generate a processing behavior representation vector; The processing behavior representation vector is processed by a fully connected layer to obtain a value between 0 and 1 as the initial processing probability. Obtain the vector of external influencing factors within the target time period, and fuse and correct the vector of external influencing factors with the preliminary processing probability to generate the business processing probability corresponding to the business data in each preset business within the target time period.

8. A business data intelligent sorting device, characterized in that, The device includes: The feature extraction module is used to acquire business data corresponding to multiple preset business, extract the structured features of the business data in each preset business, and extract the unstructured features of the business data in each preset business. The feature fusion module is used to perform feature fusion on the structured and unstructured features of the business data in each preset service using a preset cross-modal attention mechanism, to obtain the target business features of the business data in each preset service, and to convert each target business feature into a business feature vector of the business data in each preset service. The branch acquisition module is used to acquire a multi-branch neural network, which includes a first network branch for viewing probability analysis, a second network branch for demand intent analysis, and a third network branch for business processing probability analysis. The probability analysis module is used to obtain the historical viewing probability of the target personnel for each preset business data within a preset historical time period. The first network branch is used to perform a viewing probability analysis on the business data of each preset business based on the historical time period and the historical viewing probability, and to predict the viewing probability of the business data of each preset business within the target time period. The demand intent analysis module is used to perform demand intent analysis on the business feature vector using the second network branch, and generate business demand intent tags for business data in each preset business. The probability analysis module is used to perform business processing probability analysis on the business data in each preset business based on the business feature vector using the third network branch, and to predict the business processing probability of the business data in each preset business within the target time period. The priority calculation module is used to calculate the comprehensive priority feature value of business data in each preset business based on the viewing probability, the business demand intent, and the business processing probability. The initial business sorting module sorts all the business data in descending order according to the comprehensive priority feature value to generate an initial business sequence. The priority correction module is used to push the initial service sequence to the user terminal and receive real-time feedback data returned by the user terminal, and correct the comprehensive priority feature value of the service data in each preset service based on the real-time feedback data. The business target sorting module is used to reorder all the business data according to the corrected comprehensive priority feature value to generate a target business sequence.

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 business data intelligent sorting 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 intelligent sorting method for business data as described in any one of claims 1 to 7.

Images