Intention prediction method and apparatus, storage medium, and electronic device

Abstract

An intention prediction method, comprising: obtaining unique identification information of a user, network behavior data, and network performance data (S210); dividing the network behavior data and the network performance data into a plurality of dimensions, and merging network behavior data and network performance data of different dimensions, so as to obtain initial feature data (S220); performing feature fusion on the initial feature data, so as to obtain intermediate feature data (S230); processing the intermediate feature data by using a preset depth model, so as to obtain user level information and target feature data (S240); and processing the unique identification information of the user, the user level information, and the target feature data by using an intention prediction model, so as to obtain an intention prediction result (S250). The method enhances the accuracy of intention prediction and further improves the efficiency of intention prediction.

Description

Intention prediction method, device, storage medium and electronic device

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This disclosure claims priority to Chinese patent application number 202311094066.7, filed on August 28, 2023, entitled “Intention Prediction Method, Device, Storage Medium and Electronic Device,” and the entire contents of this Chinese patent application are incorporated herein by reference. Technical Field

[0003] The present disclosure relates to the field of big data technology, and in particular to an intention prediction method, an intention prediction device, a computer-readable storage medium, and an electronic device. Background Art

[0004] With the rapid development of mobile internet and the advent of the big data era, data analysis has become increasingly important. In the communications industry, in particular, telecom operators serve not only as transmission pipelines for user communication data but also as data storage providers. Scientifically and effectively utilizing data for analysis, uncovering the intent and potential value of user data, and rationally applying it to user marketing services are crucial for operators' digital transformation.

[0005] The intention prediction methods in related technologies cannot meet the current application scenarios with huge amounts of user data, which makes the user intention prediction process slow and inaccurate, resulting in low efficiency in user intention prediction.

[0006] Summary of the Invention

[0007] The present disclosure provides an intention prediction method, an intention prediction device, a computer-readable storage medium, and an electronic device, thereby improving the problem of low efficiency of intention prediction at least to a certain extent.

[0008] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by practice of the present disclosure.

[0009] According to a first aspect of the present disclosure, a method for intention prediction is provided, including: obtaining a user's unique identification information, network behavior data, and network performance data; dividing the network behavior data and network performance data into multiple dimensions, and merging the network behavior data and network performance data of different dimensions to obtain initial feature data; performing feature fusion on the initial feature data to obtain intermediate feature data; processing the intermediate feature data using a preset deep model to obtain user level information and target feature data; and processing the user's unique identification information, user level information, and target feature data using an intention prediction model to obtain an intention prediction result.

[0010] Optionally, dividing the network behavior data and network performance data into multiple dimensions, and merging the network behavior data and network performance data of different dimensions to obtain initial feature data, includes: determining M data samples in the network behavior data and network performance data, and dividing each data sample into N dimensions, and in the same data sample, merging the network behavior data and network performance data of the N dimensions to obtain M intermediate dimensional data, where M and N are positive integers; determining eigenvalues ​​and eigenvectors based on the M intermediate dimensional data, and accumulating the eigenvectors corresponding to the eigenvalues ​​to obtain the initial feature data.

[0011] Optionally, the feature fusion of the initial feature data to obtain intermediate feature data includes: feature combining the initial feature data to obtain combined feature data; classifying the combined feature data according to the similarity between the combined feature data to obtain feature classification data; and feature fusing the feature classification data to obtain the intermediate feature data.

[0012] Optionally, the feature fusion of the feature classification data to obtain the intermediate feature data includes: dividing the feature classification data into T parts, selecting any one of the T parts of the feature classification data as a test set, and using the remaining T-1 parts of the feature classification data as a training set to train a preset feature fusion model, where T is a positive integer; and determining the intermediate feature data based on T output results of the preset feature fusion model.

[0013] Optionally, the intermediate feature data is processed using a preset depth model to obtain user level information and target feature data, including: extracting the target feature data from the intermediate feature data; and processing the target feature data using the preset depth model to determine the user level information corresponding to the target feature data.

[0014] Optionally, the intention prediction model is used to process the user's unique identification information, user level information and target feature data to obtain an intention prediction result, including: one-to-one correspondence between the unique identification information, user level information, and target feature data of the same user to obtain intention recognition data; and inputting the intention recognition data into the intention prediction model to obtain the user's intention prediction result.

[0015] Optionally, after obtaining the user's unique identification information, network behavior data, and network performance data, the method further includes: performing data preprocessing on the network behavior data and network performance data.

[0016] According to the second aspect of the present disclosure, there is provided an intention prediction device, comprising: a data acquisition module, configured to acquire a user's unique identification information, network behavior data, and network performance data; a dimensionality reduction processing module, configured to divide the network behavior data and network performance data into multiple dimensions, and merge the network behavior data and network performance data of different dimensions to obtain initial feature data; a feature fusion module, configured to perform feature fusion on the initial feature data to obtain intermediate feature data; a user level information acquisition module, configured to process the intermediate feature data using a preset deep model to obtain user level information and target feature data; and an intention prediction result acquisition module, configured to process the user's unique identification information, user level information, and target feature data using an intention prediction model to obtain an intention prediction result.

[0017] According to a third aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the intention prediction method of the first aspect and its possible implementation methods are implemented.

[0018] According to a fourth aspect of the present disclosure, an electronic device is provided, comprising: a processor; and a memory for storing executable instructions of the processor, wherein the processor is configured to execute the intention prediction method of the first aspect and possible implementation thereof by executing the executable instructions.

[0019] The technical solution disclosed in this disclosure has the following beneficial effects:

[0020] On the one hand, by performing dimensionality reduction and feature fusion on user data, this approach, to a certain extent, avoids the complex model training process and data overfitting that often plague intent prediction methods in related technologies when processing massive amounts of user data. This accelerates both the model training process and the intent prediction process. Furthermore, user level information is obtained based on the intermediate feature data after feature processing, and user intent is predicted based on this user level information and the corresponding target feature data. This reduces the impact of redundant data features on intent prediction results while improving the accuracy of user intent prediction, thereby significantly increasing intent prediction efficiency. On the other hand, by acquiring user data from multiple dimensions, including unique user identification information, network behavior data, and network performance data, the richness of user data is increased, further ensuring the accuracy of intent prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] FIG1 shows a system operation architecture of this exemplary embodiment;

[0022] FIG2 shows a flow chart of an intention prediction method in this exemplary embodiment;

[0023] FIG3 shows a flow chart of obtaining initial feature data in this exemplary embodiment;

[0024] FIG4 shows a flow chart of obtaining intermediate feature data in this exemplary embodiment;

[0025] FIG5 is a flowchart showing a feature fusion process in this exemplary embodiment;

[0026] FIG6 shows a flow chart of obtaining user level information in this exemplary embodiment;

[0027] FIG7 shows a flow chart of obtaining intention prediction results in this exemplary embodiment;

[0028] FIG8 shows an architecture diagram of an intention prediction unit in this exemplary embodiment;

[0029] FIG9 shows a flowchart of another intention prediction method in this exemplary embodiment;

[0030] FIG10 is a schematic structural diagram of an intention prediction device according to this exemplary embodiment;

[0031] FIG. 11 is a schematic structural diagram of an electronic device according to this exemplary embodiment. DETAILED DESCRIPTION

[0032] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present disclosure will be more comprehensive and complete and will fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or that other methods, components, devices, steps, etc. may be employed. In other cases, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0033] In addition, the accompanying drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. Identical reference numerals in the figures denote identical or similar parts, and thus repetitive descriptions thereof will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities that do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0034] In related technologies, by obtaining historical user call records, extracting features, and training intelligent dialogue models, the intelligent dialogue models can adapt to user intention recognition in different application scenarios; or by collecting and analyzing user behavior data; interpreting user behavior data based on industry knowledge graphs, and obtaining user industry knowledge lists; applying multiple user feature recognition models to identify and output users' personalized features and needs, making the intention recognition scenarios and feature processing methods relatively simple, and not considering the processing methods of massive feature data, resulting in the intention prediction method being unable to meet the current application scenarios with huge amounts of user data, and the user intention prediction process is slow and inaccurate, resulting in low efficiency in user intention prediction.

[0035] In view of one or more of the above problems, the exemplary embodiment of the present disclosure first provides an intention prediction method. The system architecture of the operating environment of the exemplary embodiment will be described below with reference to FIG1 .

[0036] As shown in reference figure 1, the system architecture 100 may include a terminal device 110 and a server 120. The terminal device 110 may be an electronic device such as a tablet computer, a laptop computer, or a desktop computer, and the terminal device 110 may be used for the user's unique identification information, network behavior data, and network performance data. The server 120 generally refers to a background system that provides intention prediction-related services in this exemplary embodiment, such as a server that implements the intention prediction method. The server 120 may be a single server or a cluster formed by multiple servers, which is not limited in this disclosure. The terminal device 110 and the server 120 may be connected via a wired or wireless communication link for data exchange.

[0037] The intention prediction method of this exemplary embodiment can be executed by terminal device 110. For example, in a communication service scenario, terminal device 110 can be an electronic device used by a user. By executing the intention prediction method, terminal device 110 can collect the user's unique identification information, network behavior data, and network performance data, determine user level information, and thus predict the user's intention, understand the user's communication service needs, and make targeted recommendations to the user to improve the user experience.

[0038] In one embodiment, the terminal device 110 can obtain the user's unique identification information, network behavior data, and network performance data, and send the user's unique identification information, network behavior data, and network performance data to the server 120. After receiving the user's unique identification information, network behavior data, and network performance data, the server 120 divides the network behavior data and network performance data into multiple dimensions, and merges the network behavior data and network performance data of different dimensions to obtain initial feature data, and then performs feature fusion on the initial feature data to obtain intermediate feature data, and uses a preset deep model to process the intermediate feature data to obtain user level information and target feature data, and finally uses the intention prediction model to process the user's unique identification information, user level information, and target feature data to obtain an intention prediction result.

[0039] As can be seen from the above, the intention prediction method in this exemplary embodiment can be executed by the terminal device 110 or the server 120 mentioned above.

[0040] The intention prediction method is described below with reference to FIG2 . FIG2 shows an exemplary process of the intention prediction method, including the following steps S210 to S250 :

[0041] Step S210, obtaining the user's unique identification information, network behavior data, and network performance data;

[0042] Step S220: dividing the network behavior data and the network performance data into multiple dimensions, and merging the network behavior data and the network performance data of different dimensions to obtain initial feature data;

[0043] Step S230, performing feature fusion on the initial feature data to obtain intermediate feature data;

[0044] Step S240, using a preset depth model to process the intermediate feature data to obtain user level information and target feature data;

[0045] Step S250: Use the intention prediction model to process the user's unique identification information, user level information and target feature data to obtain an intention prediction result.

[0046] Based on the above method, on the one hand, by performing dimensionality reduction processing and feature fusion on user data, the problems of complex model training and data overfitting that arise in related intent prediction methods when processing massive amounts of user data are avoided to a certain extent, thereby accelerating the model training process and the intent prediction process. At the same time, user level information is obtained based on the intermediate feature data after feature processing, and user intent is predicted based on the user level information and the corresponding target feature data. This reduces the impact of redundant data features on the intent prediction results while improving the accuracy of user intent prediction, thereby greatly improving the efficiency of intent prediction. On the other hand, by obtaining user data in multiple dimensions such as user unique identification information, network behavior data, and network performance data, the richness of user data is increased, further ensuring the accuracy of intent prediction results.

[0047] Each step in FIG2 is described in detail below.

[0048] 2 , in step S210 , the user's unique identification information, network behavior data, and network performance data are obtained.

[0049] Among them, unique identification information is used to represent the user's identity information. This disclosure does not specifically limit the specific content of unique identification information. For example, unique identification information may include the user's account number or the IP (Internet Protocol) address of the terminal device used by the user. Network behavior data may include the user's network operation data. This disclosure does not specifically limit the specific content of network behavior data. For example, network behavior data may include the user's search data, communication service consumption data, etc. Network performance data may represent the network status information of the wireless network or mobile network used by the user. For example, network performance data may include data such as the bandwidth, latency, and bandwidth-delay product of the network currently used by the user.

[0050] For example, a data collection cycle may be set to regularly collect the user's unique identification information, network behavior data, and network performance data according to the data collection cycle to obtain the user feature database information table shown in Table 1.

[0051] Table 1 User feature database information table

[0052] In order to make the acquired network behavior data and network performance data easier to train in a deep model, in one embodiment, after acquiring the user's unique identification information, network behavior data, and network performance data, the above method may further include:

[0053] Perform data preprocessing on network behavior data and network performance data.

[0054] For example, data preprocessing of network behavior data and network performance data can mainly include data cleaning, data conversion, data screening, etc. In the scenario of predicting users' communication service intentions, due to the wide range of operators' user groups, the corresponding network behavior data and network performance data are also very large. Therefore, in order to more effectively extract the service characteristics of user data and predict user intentions, it is necessary to preprocess the obtained network behavior data and network performance data according to the following steps:

[0055] (1) Data import: SQL Developer, a data migration tool provided by Oracle Database, can be used to directly import the acquired user's unique identification information, network behavior data, and network performance data into the database, and the user log data in the data storage file can be exported for data analysis;

[0056] (2) Data cleaning: Since the unique identification information, network behavior data and network performance data of users are inconsistent due to data characteristics, there will be problems such as missing data, redundancy, incorrectly formatted data or dirty data. Therefore, it is necessary to clean the data for missing data, erroneous data, logical errors, redundant data and duplicate data to improve data quality. It is also convenient to improve the accuracy of the model in the subsequent user business data feature extraction step and delete the data that is of no value or not needed.

[0057] (3) Data conversion: The range normalization method can be used to normalize the data to eliminate the influence of the variable dimension and variation range on the data. For example, the maximum value (X max ) and minimum value (X min ), and calculate the range R = X min -X max , and calculate the standardized value of network behavior data or network performance data according to formula (1):

[0058] X is a sample value of the network behavior data or the network performance data, and X' is a normalized value of the network behavior data or the network performance data.

[0059] (4) Data filtering: According to formula (2), data feature screening can be performed using the variance selection method:

[0060] Where σ represents the overall variance of the data, X is the sample value of the network behavior data or network performance data, α is the overall mean of the data, and N is the number of samples of the network behavior data or network performance data.

[0061] By preprocessing the network behavior data and network performance data, the data redundancy of the processed network behavior data and network performance data is reduced, further accelerating the model convergence rate and the intention prediction process.

[0062] Continuing to refer to FIG. 2 , in step S220 , the network behavior data and the network performance data are divided into multiple dimensions, and the network behavior data and the network performance data of different dimensions are merged to obtain initial feature data.

[0063] Among them, the initial feature data includes low-dimensional data corresponding to the network behavior data and the network performance data. For example, the network behavior data may include questions searched by users, such as "Why is the router at home always disconnected from the Internet?", and the initial feature data corresponding to the network behavior data may be a set of binary digits, such as "00110101". The binary digits "00110101" can be recognized by the device as "Why is the router at home always disconnected from the Internet?", thereby obtaining low-dimensional data corresponding to the network behavior data and the network performance data.

[0064] In one embodiment, the above-mentioned network behavior data and network performance data are divided into multiple dimensions, and the network behavior data and network performance data of different dimensions are merged to obtain initial feature data. Referring to FIG. 3 , the steps S310 to S320 may be included:

[0065] Step S310: determining M data samples from the network behavior data and the network performance data, and dividing each data sample into N dimensions. Within the same data sample, merging the N dimensions of network behavior data and network performance data to obtain M intermediate dimensional data, where M and N are positive integers.

[0066] Step S320 , determining eigenvalues ​​and eigenvectors based on the M intermediate dimensional data, and accumulating the eigenvectors corresponding to the eigenvalues ​​to obtain initial eigendata.

[0067] The data samples may represent groups of data, and the M data samples may include dividing network behavior data and network performance data into M groups of data. The intermediate dimension data may be the result of merging N dimensional data in each data sample.

[0068] For example, the principal component analysis (PCA) method can be used to perform dimensionality reduction processing on network behavior data and network performance data to obtain initial feature data. For example, X is network behavior data or network performance data, X contains M data samples, and each data sample includes data of N dimensions. As shown in the following formula (3), the data of different dimensions in each data sample in X are merged to obtain formula (4).

[0069] Among them, X NM It can represent the network behavior data or network performance data of the Nth dimension in the Mth data sample. X=[X1,X2,...X M ] (4)

[0070] Among them, X M It can represent the Mth intermediate dimension data.

[0071] Then, the corresponding eigenvalues ​​and eigenvectors are calculated based on the intermediate dimensional data, the eigenvalues ​​are sorted from large to small, and the dimensionality reduction number is determined based on the cumulative contribution rate according to the following formula (5):

[0072] Among them, P k is the cumulative contribution rate, λ k Indicates the contribution rate corresponding to the kth eigenvalue.

[0073] In P k When a preset contribution rate threshold is met, accumulation can be stopped to indicate the end of dimensionality reduction, and initial feature data can be obtained based on the data of the current dimension. The preset contribution rate threshold can be 0.85.

[0074] By partitioning and merging network behavior data and network performance data to obtain low-dimensional initial feature data, we can effectively alleviate the dimensionality curse problem, reduce the complexity of the data structure, and help improve the efficiency of intent prediction.

[0075] Continuing to refer to FIG. 2 , in step S230 , feature fusion is performed on the initial feature data to obtain intermediate feature data.

[0076] The intermediate feature data is the feature fusion result of the initial feature data. In one embodiment, the feature fusion of the initial feature data to obtain the intermediate feature data may include steps S410 to S430 as shown in FIG4 :

[0077] Step S410, performing feature combination on the initial feature data to obtain combined feature data;

[0078] The combined feature data may include a new feature group formed by feature extraction and feature cross-linking based on the initial feature data. For example, if the initial feature data includes network operation data A, search behavior data B, and network performance data C, then feature extraction and feature cross-linking can be performed on the initial feature data to obtain intermediate feature data D, where D = A + C. Thus, the related initial feature data are associated through feature cross-linking to form new feature data.

[0079] Step S420 , classifying the combined feature data according to the similarity between the combined feature data to obtain feature classification data.

[0080] The similarity distance may represent the similarity between feature data. The present disclosure does not specifically limit the specific content of the similarity distance. For example, the similarity distance may include Hamming distance, Euclidean distance, etc. The feature classification data may be the result of reclassifying the combined feature data, and the similarity between the data in each category in the feature classification data is relatively high.

[0081] For example, the user's unique identification information user_id can be used as the primary key, and the combined feature data can be classified according to the similarity between the combined feature data using the K-Nearest Neighbor (KNN) algorithm or the K-means clustering algorithm to obtain feature classification data:

[0082] (1) KNN algorithm

[0083] Calculate x according to the following formula (6): k with y k The Euclidean distance d(x,y) between them is used to obtain the sample point x k The k closest neighboring points y k :

[0084] Among them, n is the total number of sample points.

[0085] In determining x k The closest k neighboring points y k After that, you can k The category with the highest frequency is determined as the sample point x k Category.

[0086] (2) K-means algorithm

[0087] If the sample set L, where L = {X1, X2, ...X n}, including n sample data X i(0<i≤n), select n cluster centers K1, K2, K3, ...K n , calculate the relationship between each cluster center point and sample data X according to formula (7) i The Euclidean distance between:

[0088] For each sample point X i Divide it into the cluster center point K closest to it i In the cluster represented by, the center point K of all sample points in each cluster is used it (0<t≤n) represents the center point of the cluster; repeat the above steps until the center point of the cluster remains unchanged or reaches the set number of iterations or the preset error tolerance range.

[0089] By classifying the combined feature data based on the similarity between the combined feature data, the intrinsic connection between the combined feature data can be mined, so that the feature classification data in each category obtained has a strong correlation.

[0090] Step S430: performing feature fusion on the feature classification data to obtain intermediate feature data.

[0091] In one embodiment, the feature fusion of the feature classification data to obtain intermediate feature data may include steps S510 to S520, as shown in FIG5 :

[0092] Step S510: Divide the feature classification data into T parts, select any one of the T parts of feature classification data as a test set, and use the remaining T-1 parts of feature classification data as a training set to train a preset feature fusion model, where T is a positive integer;

[0093] Step S520: determining intermediate feature data based on T output results of the preset feature fusion model.

[0094] The training set may be data used to train a preset feature fusion model, and the test set may be data used to verify the processing results of the preset feature fusion model. The preset feature fusion model may be a deep model used to perform feature fusion on feature classification data. The present disclosure does not specifically limit the specific content of the preset feature fusion model. Exemplarily, the preset feature fusion model may include a deep model based on the Stacking algorithm.

[0095] For example, the preset feature fusion model can be a deep model based on the improved Stacking algorithm, and the feature classification data can be divided into 5 equal parts to obtain five groups of feature classification data, namely Train1, Train2, Train3, Train4, and Train5. One of the groups of feature classification data is selected as the test set, and the remaining four groups are selected as training sets to train the preset feature fusion model to obtain five groups of prediction results Pre1, Pre2, Pre3, Pre4, and Pre5 (that is, the first-layer network output results of the preset feature fusion model); then the weighted average of the five groups of prediction results Pre1, Pre2, Pre3, Pre4, and Pre5 is calculated, and the weighted average is input into the second-layer network of the preset feature fusion model, and the intermediate feature data is obtained according to the output results of the second-layer network.

[0096] Based on the method shown in Figure 5, feature fusion is performed on the feature classification data to obtain intermediate feature data, which can further improve the accuracy and robustness of the model classification, thereby helping to improve the accuracy of intent prediction.

[0097] Based on the method in Figure 4, the initial feature data is processed through multiple models to obtain intermediate feature data. Compared with the method of using a single model for feature construction, this effectively improves the classification accuracy of the intermediate feature data and further reduces the amount of redundant data in the intermediate feature data, which is conducive to improving the overall efficiency of intent prediction.

[0098] Continuing to refer to FIG. 2 , in step S240 , the intermediate feature data is processed using a preset depth model to obtain user level information and target feature data.

[0099] Among them, the preset deep model is a neural network model based on the attention mechanism. For example, the preset deep model may include a Transformer model. User level information can be used to represent the level of the user. For example, the user level may include three levels: premium users, advanced users, and ordinary users. The target feature data may be data used to determine the user level information. The present disclosure does not specifically limit the target feature data. For example, the target feature data may be data such as the number of user visits, user visit duration, user visit frequency, user Internet traffic, and total duration of user calls extracted from the intermediate feature data.

[0100] In one embodiment, the above-mentioned processing of the intermediate feature data using the preset depth model to obtain user level information and target feature data may include steps S610 to S620 as shown in FIG6 :

[0101] Step S610, extracting target feature data from the intermediate feature data;

[0102] Step S620 : Process the target feature data using a preset depth model to determine user level information corresponding to the target feature data.

[0103] For example, the preset deep model may include an improved Transformer model, which may first extract data such as user visits, user visit duration, user visit frequency, user Internet traffic, and total user call duration from the intermediate feature data to obtain target feature data. The content description of the target feature data may be as shown in Table 2. 80% of the target feature data is used as a training set to train the preset deep model, and the remaining 20% ​​of the target feature data is used as a test set to verify the prediction accuracy of the preset deep model. The recall rate and precision rate of the model may be calculated by formula (8) to evaluate the prediction effect of the model. For example, the prediction accuracy of the model calculated based on the recall rate and precision rate of the model is 74.37%: F1 = (2*precision*recall) / (precision+recall) (8)

[0104] Among them, F1 is the prediction accuracy of the model, precision is the search accuracy, and recall is the search recall.

[0105] After the preset deep model training is completed, word vectors can be extracted from the target feature data, and the word vectors can be input into the preset deep model as initialization values ​​to learn global information. Finally, the user level information corresponding to each user's user_id is determined based on the fully connected layer and logistic regression method of the preset deep model.

[0106] Table 2 Target feature data content description

[0107] Based on the method of FIG6 , user level information corresponding to the target feature data can be obtained, user level classification is achieved, the accuracy of the obtained user level information is improved, and the accuracy of intent prediction is further improved.

[0108] Continuing to refer to FIG. 2 , in step S250 , the user's unique identification information, user level information, and target feature data are processed using the intention prediction model to obtain an intention prediction result.

[0109] Among them, the intent prediction model is a deep model used to identify user intent. The present disclosure does not specifically limit the specific type of the intent prediction model. Exemplarily, the intent prediction model may include a Text-CNN (Text Convolutional Neural Networks) model.

[0110] In one embodiment, the above-mentioned intention prediction model processes the user's unique identification information, user level information, and target feature data to obtain an intention prediction result. Referring to FIG. 7 , steps S710 to S720 may be performed:

[0111] Step S710: Match the unique identification information, user level information, and target feature data of the same user one-to-one to obtain intent recognition data;

[0112] Step S720: input the intention recognition data into the intention prediction model to obtain the user's intention prediction result.

[0113] The intent recognition data can be used to identify user intent. The intent prediction result can include the output of the intent prediction model. This disclosure does not specifically limit the specific content of the intent prediction result. For example, the intent prediction result can include the user's communication service needs and the user's basic information.

[0114] For example, the Text-CNN model is used to encode the target feature data and concatenate it with the feature word vector to form a word vector with contextual feature information. The probability distribution result of the intent category label is then obtained based on the word vector, and finally the user's intent prediction result is determined based on the probability distribution result.

[0115] In one embodiment, the intention prediction method of this exemplary embodiment may be integrated to obtain an intention prediction unit, and the intention prediction unit may be embedded in a communication system to predict the user's communication service needs.

[0116] For example, the architecture of the intention prediction unit can be shown in Figure 8. The intention prediction unit can include a data management subunit, a data acquisition subunit, a feature engineering subunit, a model training subunit and an intention prediction subunit. The network data of all users can be stored in the data storage subunit. In the data acquisition subunit, user basic data including the user's unique identification information, network operation data, search data and network performance data can be screened in the network data. After obtaining these data, they can be cleaned, preprocessed, feature selected (data dimension reduction, data classification) and feature combined in the feature engineering subunit to obtain initial feature data. The initial feature data is input into the model training subunit to perform model training on the preset feature fusion model and the preset depth model. The initial feature data is feature fused based on the trained preset feature fusion model to obtain To the intermediate feature data, the intermediate feature data is input into the trained preset deep model to obtain user level information; in the intention prediction subunit, the user level information, the user's unique identification information and the target feature data are input into the intention prediction model to obtain the user's intention prediction result, based on the user's intention prediction result, the user's business demand information is determined, and the service strategy is formulated according to the business demand information. The formulated service strategy is sent to the business processing subunit, so that the business processing subunit can perform marketing services, operation scheduling, user management and fault alarm according to the service strategy, and update the user's network data, and at the same time, the updated data is stored again in the data management subunit.

[0117] By integrating the intention prediction method into the intention prediction unit and embedding the intention prediction unit into the communication service system, the communication service system can predict the user's intention based on the user's network data, thereby providing the possibility for the communication service system to perform business intelligent processing.

[0118] In one embodiment, referring to FIG9 , an exemplary process of the intention prediction method of this exemplary embodiment may include steps S901 to S909:

[0119] Step S901, obtaining user's network data;

[0120] Step S902: Acquire the user's unique identification information, network behavior data, and network performance data based on the user's network data;

[0121] Step S903: pre-processing the network behavior data and network performance data;

[0122] Step S904: performing data dimensionality reduction on the pre-processed network behavior data and network performance data based on the PCA method to obtain initial feature data;

[0123] Step S905, performing feature combination on the initial feature data to obtain combined feature data;

[0124] Step S906, using the KNN algorithm and the K-means algorithm to classify the combined feature data to obtain feature classification data;

[0125] Step S907: performing feature fusion on the feature classification data based on the Stacking algorithm model to obtain intermediate feature data;

[0126] Step S908: Process the intermediate feature data based on the Transformer model to obtain user level information and target feature data;

[0127] Step S909: Use the intention prediction model to process the user's unique identification information, user level information and target feature data to obtain an intention prediction result.

[0128] The method based on Figure 9 effectively improves the accuracy of intent prediction and further improves the efficiency of intent prediction.

[0129] Based on the above method, the problems of complex model training process and data overfitting that arise in the intention prediction methods in related technologies when processing massive user data are avoided to a certain extent, the model training process and the intention prediction process are accelerated, the accuracy of user intention prediction is improved, and thus the efficiency of intention prediction is greatly improved.

[0130] The exemplary embodiment of the present disclosure further provides an intention prediction device. As shown in FIG10 , the intention prediction device 1000 may include:

[0131] The data acquisition module 1010 is configured to acquire the user's unique identification information, network behavior data, and network performance data;

[0132] Dimensionality reduction processing module 1020 is configured to divide the network behavior data and the network performance data into multiple dimensions, and merge the network behavior data and the network performance data of different dimensions to obtain initial feature data;

[0133] The feature fusion module 1030 is configured to perform feature fusion on the initial feature data to obtain intermediate feature data;

[0134] The user level information acquisition module 1040 is configured to process the intermediate feature data using a preset depth model to obtain user level information and target feature data;

[0135] The intention prediction result acquisition module 1050 is configured to process the user's unique identification information, user level information and target feature data using the intention prediction model to obtain the intention prediction result.

[0136] In one embodiment, the above-mentioned dividing the network behavior data and the network performance data into multiple dimensions and merging the network behavior data and the network performance data of different dimensions to obtain the initial feature data may include:

[0137] Determine M data samples from the network behavior data and the network performance data, and divide each data sample into N dimensions. In the same data sample, merge the network behavior data and the network performance data of the N dimensions to obtain M intermediate dimension data, where M and N are positive integers.

[0138] Determine the eigenvalues ​​and eigenvectors based on the M intermediate dimensional data, and accumulate the eigenvectors corresponding to the eigenvalues ​​to obtain initial feature data.

[0139] In one embodiment, the feature fusion of the initial feature data to obtain the intermediate feature data may include:

[0140] Performing feature combination on the initial feature data to obtain combined feature data;

[0141] Classifying the combined feature data according to the similarity between the combined feature data to obtain feature classification data;

[0142] Perform feature fusion on feature classification data to obtain intermediate feature data.

[0143] In one embodiment, the above-mentioned feature fusion of the feature classification data to obtain intermediate feature data may include:

[0144] Divide the feature classification data into T parts, select any one of the T parts as the test set, and use the remaining T-1 parts as the training set to train the preset feature fusion model, where T is a positive integer;

[0145] Determine the intermediate feature data based on the T-times output results of the preset feature fusion model.

[0146] In one embodiment, the above-mentioned processing of the intermediate feature data using the preset depth model to obtain the user level information and the target feature data may include:

[0147] Extracting target feature data from intermediate feature data;

[0148] The target feature data is processed using a preset depth model to determine the user level information corresponding to the target feature data.

[0149] In one embodiment, the above-mentioned processing of the user's unique identification information, user level information, and target feature data using the intention prediction model to obtain the intention prediction result may include:

[0150] Match the unique identification information, user level information, and target feature data of the same user one by one to obtain intent recognition data;

[0151] The intent recognition data is input into the intent prediction model to obtain the user's intent prediction results.

[0152] In one embodiment, after obtaining the user's unique identification information, network behavior data, and network performance data, the apparatus may further include:

[0153] Perform data preprocessing on network behavior data and network performance data.

[0154] The specific details of each part of the above device have been described in detail in the method part of the implementation method, so they will not be repeated here.

[0155] Based on the above-mentioned device, on the one hand, by performing dimensionality reduction processing and feature fusion on user data, the problems of complex model training process and data overfitting that arise in the intention prediction methods in related technologies when processing massive amounts of user data are avoided to a certain extent, thereby accelerating the model training process and the intention prediction process. At the same time, user level information is obtained based on the intermediate feature data after feature processing, and user intention is predicted based on the user level information and the corresponding target feature data. While reducing the impact of redundant data features on the intention prediction results, the accuracy of user intention prediction is improved, thereby greatly improving the efficiency of intention prediction. On the other hand, by obtaining user data in multiple dimensions such as user unique identification information, network behavior data, and network performance data, the richness of user data is improved, further ensuring the accuracy of intention prediction results.

[0156] The exemplary embodiments of the present disclosure also provide a computer-readable storage medium that can be implemented in the form of a program product, which includes program code. When the program product is run on an electronic device, the program code is used to cause the electronic device to perform the steps described in the "Exemplary Method" section above of this specification according to the various exemplary embodiments of the present disclosure. In an optional embodiment, the program product can be implemented as a portable compact disc read-only memory (CD-ROM) and include program code, and can be run on an electronic device, such as a personal computer. However, the program product of the present disclosure is not limited to this, and in this document, a readable storage medium can be any tangible medium that contains or stores a program that can be used by or in combination with an instruction execution system, apparatus, or device.

[0157] The program product may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination thereof. More specific examples (a non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0158] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0159] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0160] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, and the like, as well as conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a stand-alone software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0161] The exemplary embodiments of the present disclosure further provide an electronic device. The electronic device may include a processor and a memory. The memory stores executable instructions for the processor, such as program code. The processor executes the executable instructions to perform the method of the exemplary embodiment.

[0162] 11 , an electronic device is exemplarily described in the form of a general-purpose computing device. It should be understood that the electronic device 1100 shown in FIG11 is merely an example and should not limit the functionality and scope of use of the embodiments of the present disclosure.

[0163] As shown in FIG. 11 , the electronic device 1100 may include a processor 1110 , a memory 1120 , a bus 1130 , an I / O (input / output) interface 1140 , and a network adapter 1150 .

[0164] The processor 1110 may include one or more processing units, for example: the processor 1110 may include a central processing unit (CPU), an application processor (AP), a modem processor, a display processor (DPU), a graphics processing unit (GPU), an image signal processor (ISP), a controller, an encoder, a decoder, a digital signal processor (DSP), a baseband processor, an artificial intelligence processor, etc. In one embodiment, the GPU may obtain the user's unique identification information, network behavior data, and network performance data; divide the network behavior data and network performance data into multiple dimensions, merge the network behavior data and network performance data of different dimensions to obtain initial feature data; then perform feature fusion on the initial feature data to obtain intermediate feature data; use a preset deep model to process the intermediate feature data to obtain user level information and target feature data; and finally use the intent prediction model to process the user's unique identification information, user level information, and target feature data to obtain an intent prediction result.

[0165] The memory 1120 may include volatile memory, such as RAM 1121 and cache unit 1122, and may also include non-volatile memory, such as ROM 1123. The memory 1120 may also include one or more program modules 1124. Such program modules 1124 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. For example, the program modules 1124 may include the modules in the aforementioned apparatus 1000.

[0166] The bus 1130 is used to realize the connection between different components of the electronic device 1100 and may include a data bus, an address bus, and a control bus.

[0167] The electronic device 1100 can communicate with one or more external devices 1200 (eg, a keyboard, a mouse, an external controller, etc.) through the I / O interface 1140 .

[0168] The electronic device 1100 can communicate with one or more networks via a network adapter 1150. For example, the network adapter 1150 can provide mobile communication solutions such as 3G / 4G / 5G, or wireless communication solutions such as wireless LAN, Bluetooth, and near-field communication. The network adapter 1150 can communicate with other modules of the electronic device 1100 via the bus 1130.

[0169] Although not shown in FIG. 11 , other hardware and / or software modules may be provided in the electronic device 1100 , including but not limited to: a display, microcode, device drivers, redundant processors, an external disk drive array, a RAID system, a tape drive, and a data backup storage system.

[0170] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the exemplary embodiments of the present disclosure, the features and functions of two or more modules or units described above can be concretized in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0171] It will be appreciated by those skilled in the art that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented as the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects, which may be collectively referred to herein as a "circuit", "module" or "system". Other embodiments of the present disclosure will readily occur to those skilled in the art after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to encompass any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and implementation are intended to be exemplary only, with the true scope and spirit of the present disclosure being indicated by the claims.

[0172] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. An intention prediction method, comprising: Obtain the user's unique identification information, network behavior data, and network performance data; Dividing the network behavior data and the network performance data into multiple dimensions, and merging the network behavior data and the network performance data of different dimensions to obtain initial feature data; Performing feature fusion on the initial feature data to obtain intermediate feature data; Processing the intermediate feature data using a preset deep model to obtain user level information and target feature data; The intention prediction model is used to process the user's unique identification information, user level information and target feature data to obtain an intention prediction result.

2. The method according to claim 1, wherein: The dividing the network behavior data and the network performance data into multiple dimensions, and merging the network behavior data and the network performance data of different dimensions to obtain initial feature data, includes: Determine M data samples in the network behavior data and the network performance data, and divide each of the data samples into N dimensions, and in the same data sample, merge the network behavior data and the network performance data of the N dimensions to obtain M intermediate dimensional data, where M and N are positive integers; Determine eigenvalues ​​and eigenvectors according to the M intermediate dimensional data, and accumulate the eigenvectors corresponding to the eigenvalues ​​to obtain the initial eigendata.

3. The method according to claim 1, wherein: The performing feature fusion on the initial feature data to obtain intermediate feature data includes: Performing feature combination on the initial feature data to obtain combined feature data; Classifying the combined feature data according to the similarity between the combined feature data to obtain feature classification data; Feature fusion is performed on the feature classification data to obtain the intermediate feature data.

4. The method according to claim 3, wherein: The performing feature fusion on the feature classification data to obtain the intermediate feature data includes: Divide the feature classification data into T parts, select any one of the feature classification data as a test set from the T parts of the feature classification data, and use the remaining T-1 parts of the feature classification data as a training set to train a preset feature fusion model, where T is a positive integer; The intermediate feature data is determined according to T output results of the preset feature fusion model.

5. The method according to claim 1, wherein: The using a preset depth model to process the intermediate feature data to obtain user level information and target feature data includes: Extracting the target feature data from the intermediate feature data; The target feature data is processed using the preset depth model to determine user level information corresponding to the target feature data.

6. The method according to claim 1, wherein: The use of the intention prediction model to process the unique identification information, user level information and target feature data of the user to obtain an intention prediction result includes: Matching the unique identification information of the same user, the user level information, and the target feature data one by one to obtain intent recognition data; The intention recognition data is input into the intention prediction model to obtain the intention prediction result of the user.

7. The method according to claim 1, wherein: After obtaining the user's unique identification information, network behavior data, and network performance data, the method further includes: Data preprocessing is performed on the network behavior data and the network performance data.

8. An intention prediction device, comprising: A data acquisition module is configured to acquire unique identification information, network behavior data and network performance data of a user; A dimensionality reduction processing module is configured to divide the network behavior data and the network performance data into multiple dimensions, and merge the network behavior data and the network performance data of different dimensions to obtain initial feature data; A feature fusion module is configured to perform feature fusion on the initial feature data to obtain intermediate feature data; A user level information acquisition module is configured to process the intermediate feature data using a preset deep model to obtain user level information and target feature data; The intention prediction result acquisition module is configured to use the intention prediction model to process the user's unique identification information, user level information and target feature data to obtain the intention prediction result.

9. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method according to any one of claims 1 to 7.

10. An electronic device, comprising: processor; A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 7 by executing the executable instructions.

Images