Data processing method, electronic device, and computer storage medium
By converting sample data into a fixed format and performing noise exchange, the first machine learning model is trained to extract latent vector features, which solves the problems of imbalanced samples and insufficient samples, and realizes effective training of machine learning models in customer service scenarios and improves service efficiency.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ALIBABA (CHINA) CO LTD
- Filing Date
- 2022-01-20
- Publication Date
- 2026-05-19
AI Technical Summary
In cases of imbalanced or insufficient samples, machine learning models often underfit, resulting in poor training performance. This is particularly pronounced in customer service hotline call scenarios, where there are few positive samples and many negative samples.
The sample data is converted into fixed format data, and noisy sample data is generated by exchanging noise. The first machine learning model is trained, and the output is used as the input to build the second machine learning model. Feature extraction is performed using the latent vector features of the first machine learning model, and the second machine learning model is fine-tuned by a small number of positive and negative samples.
It solves the problems of imbalanced positive and negative samples and insufficient samples, ensuring the training effect of machine learning models and improving the efficiency of consultation consumption conversion and service efficiency in service scenarios.
Smart Images
Figure CN114492839B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of computer technology, and in particular to a data processing method, electronic device, and computer storage medium. Background Technology
[0002] With the development of artificial intelligence technology, intelligent decision-making based on machine learning models has been widely used, such as intelligent recommendation and autonomous driving.
[0003] Currently, most machine learning models used for intelligent decision-making, such as LR+GBDT, Wide & Deep, DeepFM, and DIN, require a balanced mix of positive and negative samples. Simultaneously, machine learning models, represented by DeepFM, also have requirements regarding the number of samples; machine learning methods often need more samples to fit the parameters, and in cases of imbalanced or insufficient samples, the model often underfits. However, in real-world scenarios, there is often a problem of too few positive samples and too many negative samples, posing a significant challenge to achieving the performance targets of the original model. For example, in customer service hotline call scenarios, the imbalance is exacerbated by the fact that much of the hotline call data is inherently imbalanced, and since most of the data is manually registered, the sample imbalance problem is even more pronounced.
[0004] Therefore, how to ensure the training effect of machine learning models under imbalanced sample conditions, and thus obtain more accurate service information, has become an urgent problem to be solved. Summary of the Invention
[0005] In view of this, embodiments of this application provide a data processing scheme to at least partially solve the above-mentioned problems.
[0006] According to a first aspect of the embodiments of this application, a data processing method is provided, comprising: converting sample data into sample format data according to a preset fixed format; performing noise exchange processing on the sample format data to obtain noisy sample format data; constructing a first training sample based on the noisy sample format data; and training a first machine learning model for feature extraction, wherein the output of the trained first machine model is used as the input to construct a second machine learning model for service processing.
[0007] According to a second aspect of the embodiments of this application, another data processing method is provided, comprising: determining user information corresponding to the user identifier of the user to be served; inputting fixed-format data corresponding to the user information into a first machine learning model to obtain representation data output by the first machine learning model; inputting the user information and the representation data into a second machine learning model to obtain service processing data output by the second machine learning model; and providing corresponding service information to the user according to the service processing data.
[0008] According to a third aspect of the embodiments of this application, another data processing method is provided, comprising: receiving a customer service consultation call from a customer; obtaining the corresponding call content and customer information corresponding to the customer based on the customer service consultation call, and sending the call content and the customer information to a server; and receiving a marketing strategy for the customer determined by the server based on the call content and the customer information.
[0009] According to a fourth aspect of the embodiments of this application, an electronic device is provided, including: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other through the communication bus; the memory is used to store at least one executable instruction, wherein the executable instruction causes the processor to perform an operation corresponding to the method described in the first, second, or third aspect.
[0010] According to a fourth aspect of the embodiments of this application, a computer storage medium is provided that stores a computer program thereon, which, when executed by a processor, implements the method as described in the first, second, or third aspect.
[0011] According to the data processing scheme provided in the embodiments of this application, firstly, for the sample data, on the one hand, it is converted into a fixed format data form, namely sample format data, which facilitates the first machine learning model to learn the relationship between text and semantics; on the other hand, the robustness of the first machine learning model is improved by adding noise; furthermore, the noise addition is achieved by exchanging noise processing, which not only effectively achieves noise processing, but also ensures that different types of data in the sample format data do not have too much impact; furthermore, the training of the first machine learning model enables the first machine learning model to effectively learn the latent vector features of the sample data. Subsequently, the latent vector features output by the trained first machine learning model are used to construct the training samples of the traditional second machine learning model, so that the second machine learning model only needs a small number of positive and negative samples for fine-tuning to complete the training, which solves the problems of imbalanced positive and negative samples and insufficient samples, while ensuring the training effect of the traditional machine learning model.
[0012] Secondly, applying the trained machine learning model to real-world service scenarios, such as customer service, can effectively obtain relevant data from customer inquiries and, based on the machine learning model, output marketing strategies after processing this data (including call content and customer communication feature information obtained from customer information), thereby improving the efficiency of consultation and consumption conversion.
[0013] However, it is not limited to this. The trained machine learning model can also be widely applied to other applicable scenarios. In these scenarios, after obtaining the user information of the user to be served, the corresponding fixed-format data can be obtained based on it. Then, based on the processing of the first machine learning model and the second machine learning model, the data required for various scenarios, namely service processing data, can be obtained. Based on the service processing data, the corresponding service information can be provided to the user, which effectively improves service efficiency and user satisfaction. Attached Figure Description
[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.
[0015] Figure 1A This is a schematic diagram of a machine learning model architecture in an embodiment of this application;
[0016] Figure 1B This is a schematic diagram of an exemplary system for the data processing method applicable to the embodiments of this application;
[0017] Figure 2A This is a flowchart of the steps of a data processing method according to Embodiment 1 of this application;
[0018] Figure 2B for Figure 2A A schematic diagram of a model training phase in the illustrated embodiment;
[0019] Figure 3A This is a flowchart of the steps of a data processing method according to Embodiment 2 of this application;
[0020] Figure 3B for Figure 3A A schematic diagram of a model inference stage in the illustrated embodiment;
[0021] Figure 4A This is a flowchart of the steps of a data processing method according to Embodiment 3 of this application;
[0022] Figure 4B for Figure 4AA schematic diagram illustrating a scenario example of the embodiment shown;
[0023] Figure 5 This is a schematic diagram of the structure of an electronic device according to Embodiment 4 of this application. Detailed Implementation
[0024] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of this application.
[0025] To understand the data processing scheme of this application embodiment, the following first describes the machine learning model architecture used in the data processing scheme, such as... Figure 1A As shown.
[0026] The architecture includes two machine learning models: a first machine learning model and a second machine learning model. The first machine learning model, based on the input user information, performs feature extraction and other processing, outputting a corresponding intermediate feature vector. This intermediate feature vector carries rich user features extracted from the user information; in this embodiment, it is also referred to as representation data. The data input to the first machine learning model is data in a fixed format, which facilitates the effective representation of the data and the relationships between their meanings. In one feasible approach, when the first machine learning model uses a table-based model, the user information needs to be converted into tabular data before being input into the first machine learning model.
[0027] The second machine learning model is used to perform service-related processing based on the intermediate feature vectors output by the first machine learning model and the original user information, and then outputs the corresponding service processing data. The service described varies depending on the actual application, and this embodiment does not impose any limitations on it. For example, it could be a recommendation service or a quality evaluation service, etc.
[0028] Based on the output of this second machine learning model, corresponding follow-up services can be provided, such as information recommendation or service quality evaluation.
[0029] The above architecture can be deployed in the corresponding system to achieve the corresponding functions. Figure 1B An exemplary system is shown that implements the data processing method of the embodiments of this application using the above-described architecture.
[0030] like Figure 1BAs shown, the system 100 may include a server 102, a communication network 104, and / or one or more user devices 106. Figure 1B The example in the text shows multiple user devices.
[0031] Server 102 can be any suitable server for storing information, data, programs, and / or any other suitable type of content. In this embodiment, the server 102 may be equipped with the aforementioned machine learning model architecture to perform corresponding service functions. Furthermore, the training of the machine learning model in the aforementioned architecture can also be implemented within the server 102. In some embodiments, after obtaining service processing data through the deployed machine learning model architecture, server 102 can send this data to user device 106, allowing user device 106 to perform further processing according to actual needs.
[0032] In some embodiments, communication network 104 may be any suitable combination of one or more wired and / or wireless networks. For example, communication network 104 may include any one or more of the following: the Internet, intranet, wide area network (WAN), local area network (LAN), wireless network, digital subscriber line (DSL) network, frame relay network, asynchronous transfer mode (ATM) network, virtual private network (VPN), and / or any other suitable communication network. User equipment 106 may be connected to communication network 104 via one or more communication links (e.g., communication link 112), and communication network 104 may be linked to server 102 via one or more communication links (e.g., communication link 114). Communication links may be any communication link suitable for transmitting data between user equipment 106 and server 102, such as network links, dial-up links, wireless links, hardwired links, any other suitable communication links, or any suitable combination of such links.
[0033] User device 106 may include user devices suitable for presenting information, including but not limited to any one or more of text, images, audio, and video. In some embodiments, user device 106 may determine the user identifier of the user to be served and send the user identifier to server 102, so that server 102 can obtain user information based on the user identifier, perform subsequent data processing based on the user information, and finally obtain service information for the user to be served, and may feed back the service information to user device 106. In some embodiments, user device 106 may include any suitable type of device. For example, in some embodiments, user device 106 may include mobile devices, tablet computers, laptop computers, desktop computers, wearable computers, game consoles, media players, vehicle entertainment systems, and / or any other suitable type of user device.
[0034] Although server 102 is illustrated as a single device, in some embodiments, any suitable number of devices may be used to perform the functions performed by server 102. For example, in some embodiments, multiple devices may be used to implement the functions performed by server 102. Alternatively, cloud services may be used to implement the functions of server 102.
[0035] Based on the above architecture and system, this application provides a data processing scheme, which will be described below through several embodiments.
[0036] Example 1
[0037] Reference Figure 2A The diagram shows a flowchart of the steps of a data processing method according to Embodiment 1 of this application.
[0038] In this embodiment, the training of the first and second machine learning models in the above architecture is first described. The data processing method in this embodiment includes the following steps:
[0039] Step S202: Convert the sample data into sample format data according to the preset fixed format.
[0040] The fixed format may vary depending on the specific needs, as long as it effectively represents the data and the relationships between them. One feasible approach is to use a tabular format, where the data is represented in a table format. For a table format, it must have at least N columns, or N fields, where N is a positive integer.
[0041] In this embodiment, the sample data can be labeled sample data, unlabeled sample data, or a mixture of both. The specific implementation of converting the sample data into sample format data can be carried out by those skilled in the art according to actual needs, and this embodiment does not limit this. For example, based on the information in the fields of the table format, the corresponding parts of the sample data can be extracted and a correspondence established between them, or the data can be stored accordingly.
[0042] For example, suppose sample data A includes the following information: {Name: Xiaoming; Age: 23; City: Beijing; Subject: Computer; Grade: 90; City: XXX}. If this data is authorized for use, and the table format includes four fields: {Name, City, Subject, Grade}, then based on this, the information in sample data A is extracted accordingly, generating the corresponding sample table data, as shown in Table 1 below:
[0043] Name City Subject score Xiaoming Beijing computer 90
[0044] As can be seen, after converting the sample data into sample tabular data, it can be presented in tabular form from a presentation perspective. From the perspective of data relationships, there is a correspondence between the sample data and the table fields, and this correspondence can also effectively represent the semantic information of the corresponding data in the sample data. For example, "computer" is the specific implementation of "subject".
[0045] In one feasible approach, the sample data includes the following information: user attribute information and / or user behavior information. User attribute information characterizes a user's basic attributes, such as city, age, and location; user behavior information characterizes user behavior over a period of time. Optionally, this user behavior information can be service characteristic information related to a preset service. Taking telecommunications operation services as an example, the user behavior information may include at least one of the following: user behavior characteristic information on a daily basis (e.g., whether there were calls to customer service, whether a telecommunications package was purchased), user behavior characteristic information on a monthly basis (e.g., monthly behavior characteristics of the user over the past two years, such as whether they visited telecommunications operation locations, the type of telecommunications package used, etc.), and user call-in behavior characteristics (e.g., whether there were inquiries, and the details of the inquiries). This allows for a more comprehensive and accurate acquisition of user-related characteristics, providing a basis for subsequent user-related service processing.
[0046] Step S204: Perform noise exchange processing on the sample format data to obtain noisy sample format data.
[0047] Swap noise processing is used to add noise to the original data by swapping data. It can typically include column swap noise, row swap noise, and random swap noise. Column swap noise is applied to the columns of the sample table data, row swap noise is applied to the rows of the sample table data, and random swap noise adds noise by randomly swapping data in the sample table data.
[0048] In a preferred embodiment of this application, column swapping noise processing is performed on the sample table data to obtain noisy sample table data. That is, for each column of sample table data, the values at some positions are randomly swapped. For example, as shown... Figure 2B As shown in the table, and as can be seen from the figure, in the two records corresponding to "Xiaoming" and "Xiaobai", the values of the city field and the speech time field were swapped before and after the noise processing. By using column swapping for noise processing, noise can be added based on the original data, while having little impact on the meaning represented by each field in the table, which facilitates the subsequent training of the first machine learning model.
[0049] Step S206: Construct the first training sample based on the noisy sample format data, and train the first machine learning model for feature extraction.
[0050] In this embodiment, the output of the trained first machine learning model is used as the input to construct the second machine learning model for service processing. During the training phase, the output of the first machine learning model can be used to construct training samples for the second machine learning model; while during the inference phase, the output of the first machine learning model can be used to construct the input information for the second machine learning model. The second machine learning model is used to perform relevant service processing, including but not limited to: recommendation services, quality evaluation services, etc. In this embodiment, the training of the first machine learning model can be carried out using unsupervised training.
[0051] In one feasible approach, the first machine learning model can be implemented as an encoder model based on fixed-format data, such as tabular data. Since the first training samples are noisy training samples, in one feasible approach, this encoder model can be implemented as a denoising autoencoder such as a DAE, or it can also be implemented as a Transformer encoder. However, to obtain better latent vector representations, this embodiment uses a tabular-encoder with denoising capabilities based on tabular data. This tabular-encoder differs from a DAE in that it transforms the original dimensionality reduction part in the DAE into an dimensionality increase part, such as... Figure 2B The circular nodes in the middle layer of the Tabular-Encoder are shown in the diagram. This allows for a more complete representation of the data, resulting in more accurate representations.
[0052] Based on this, in one feasible approach, constructing a first training sample based on noisy sample format data and performing unsupervised training on a first machine learning model used for feature extraction can be achieved as follows: constructing the first training sample based on noisy sample format data, such as sample table data; inputting the first training sample into the input layer of the encoder (e.g., ... Figure 2B The topmost layer of the Tabular-Encoder shown; through the encoder's hidden layers (such as...) Figure 2B The intermediate layer of the Tabular-Encoder shown in the diagram increases the dimensionality of the first training samples received by the input layer and extracts features based on the increased dimensionality first training samples; the output layer of the encoder (such as...) Figure 2B The bottom layer of the Tabular-Encoder shown outputs the sample representation data corresponding to the first training sample based on the feature extraction results.
[0053] Subsequently, during the inference phase of the first machine learning model, it can directly output corresponding representation data based on fixed-format data such as tabular data of the input user information. This representation data carries rich information that represents user features extracted from the user information.
[0054] After the first machine learning model has been trained, optionally, the following steps S208-S210 can be performed.
[0055] Step S208: Obtain labeled sample data and sample representation data obtained by feature extraction of the sample format data corresponding to the labeled sample data by the first machine learning model that has been trained.
[0056] Because the input to the first machine learning model is data in a fixed format, the labeled sample data needs to be converted into the corresponding sample format data. The labels corresponding to this labeled sample data vary depending on the actual needs. For example, if the actual need is service quality assessment, the labels might include: good, medium, bad, etc.; or, if the actual need is whether to make a recommendation, the labels might include: recommendation successful, recommendation failed, etc. In this embodiment, the specific implementation of the labels is not limited.
[0057] It should be noted that the labeled sample data may be from the first training sample or it may be entirely new data from a non-first training sample. Regardless of the case, feature extraction must be performed on the data using the trained first machine learning model to obtain the sample representation data output by the first machine learning model for this iteration.
[0058] Step S210: Construct a second training sample based on the labeled sample data, the labels corresponding to the labeled sample data, and the sample representation data corresponding to the labeled sample data, and train the second machine learning model used for service processing.
[0059] In one feasible approach, unlike the first machine learning model, the second machine learning model employs supervised training. This is because the trained first machine learning model can output relatively accurate sample representation data that can characterize user features. This relatively accurate sample representation data enables the second machine learning model to train relatively accurate parameters even with a small sample size.
[0060] When training the second machine learning model, labeled sample data and its corresponding sample representation data are used as input, and the labels corresponding to the labeled sample data are used as supervision to train the second machine learning model until the training termination condition is met, such as reaching a predetermined number of training times or the loss value reaching a preset threshold.
[0061] In this embodiment, the specific form of the second machine learning model is not limited. It can output corresponding service processing data, such as quality assessment data, recommendation data, or recommendation assessment data, based on the input data. For example, the second machine learning model can be implemented as a DNN model or a CNN model, etc.
[0062] The following example illustrates the above process. Figure 2B As shown.
[0063] First, obtain sample data. In this example, only two sample data are used as an example. However, those skilled in the art should understand that in practical applications, more sample data can be processed in accordance with the sample data processing in this example.
[0064] The two sample datasets are: {Name: Xiaoming; Age: 23; City: Beijing; Data usage: 10GB, Voice calls: 100 minutes, SMS messages: 100} and {Name: Xiaobai; Age: 20; City: Shanghai; Data usage: 5GB, Voice calls: 10 minutes, SMS messages: 50}. Furthermore, assume that the table format in this example is as follows: Figure 2B As shown, it includes six fields: Name, Age, City, Data, Voice, and Message. Based on this, the converted sample table data is as follows: Figure 2B As shown in the table at the top left, in this example, this data is referred to as tabular data.
[0065] Furthermore, the sample table data converted from the sample data undergoes column-swapping noise processing. The resulting noisy sample table data is as follows: Figure 2B The table in the lower left corner is shown. Unlike traditional noise-adding and masking operations, this example randomly swaps the data in each column. Because tabular data contains different data types, the noise processing of column swapping has a smaller impact on different data types.
[0066] As can be seen, the "City" and "Voice" fields for "Xiaoming" and "Xiaobai" have been swapped. The noisy sample table data for "Xiaoming" and "Xiaobai" will be input into the first machine learning model for training. In this example, the first machine learning model is implemented as a Tabular-Encoder. The tabular data after column swapping and noise processing is input into the Tabular-Encoder for unsupervised learning. In this example, the network structure of the Tabular-Encoder differs from the traditional DAE (Denoising Auto-encoder). The original dimensionality reduction part in the DAE is improved into a dimensionality increase operation to make the data more fully represented, thereby obtaining the representation output A of the tabular data. For example, assuming "Xiaoming" is used as a training sample, the first machine learning model will train to output "representation output A". The training of the first machine learning model is iterated until the training termination condition is met, such as satisfying a preset number of training iterations.
[0067] Furthermore, such as Figure 2B As shown below the dashed line, assuming the original sample data corresponding to the representation output A is labeled sample data, after obtaining the representation output A, the labeled sample data and the representation output A can be mapped, and then input into the second machine learning model. In this example, this is illustrated as an MLP model (which can be specifically implemented as a DNN model, etc.) for training. The training of the second machine learning model is also iterative until the training termination condition is met, such as meeting a preset number of training iterations or the loss value meeting a predetermined threshold.
[0068] In this embodiment, the sample data is converted into a fixed format, namely sample format data, which facilitates the first machine learning model in learning the relationship between text and semantics. Furthermore, noise is added to improve the robustness of the first machine learning model. The noise is also achieved through a swapping noise processing method, effectively adding noise without significantly affecting different types of data in the sample format data. Finally, training the first machine learning model enables it to effectively learn the latent vector features of the sample data. The latent vector features output by the trained first machine learning model are then used to construct training samples for a traditional second machine learning model. This allows the second machine learning model to complete training with only a small number of positive and negative samples, solving the problems of imbalanced positive and negative samples and insufficient samples, while ensuring the training effect of the traditional machine learning model.
[0069] Example 2
[0070] Reference Figure 3AThe flowchart illustrates the steps of a data processing method according to Embodiment 2 of this application.
[0071] This embodiment focuses on the data processing method of this application embodiment, specifically the processing during the model inference phase. The data processing method of this embodiment includes the following steps:
[0072] Step S302: Determine the user information corresponding to the user identifier based on the user identifier of the user to be served.
[0073] In one feasible approach, user information includes at least user attribute information and / or user behavior information.
[0074] The first and second machine learning models, once trained, are used together during the inference phase to directly obtain service processing data for the user based on the user information, such as recommendation information, recommendation strategies, quality assessment data, and so on.
[0075] Users can input relevant information through their terminal devices, which may include a user identifier. Based on this user identifier, the server can obtain user information corresponding to that user, including but not limited to user attribute information and / or user behavior information. The specific content of the user attribute information and user behavior information can be referred to the descriptions in the relevant parts of the foregoing embodiments, and will not be repeated here.
[0076] Step S304: Input the fixed-format data corresponding to the user information into the first machine learning model to obtain the representation data output by the first machine learning model.
[0077] As mentioned above, the first machine learning model is an encoder model, which can optionally be an encoder model based on tabular data. Therefore, user information needs to be converted into corresponding tabular data and then input into the first machine learning model. Through the processing of the first machine learning model, the corresponding representation data is output.
[0078] Step S306: Input user information and representation data into the second machine learning model to obtain the service processing data output by the second machine learning model.
[0079] After obtaining the representation data, the representation data can be combined with user information and input into the second machine learning model. The second machine learning model will process the data and output the corresponding service processing data, including but not limited to recommendation information, recommendation evaluation data, and quality evaluation data.
[0080] Step S308: Provide the user with corresponding service information based on the service processing data.
[0081] For example, if the output of the second machine learning model is recommendation information, then recommendations are made to the user based on the recommendation information; if the output of the second machine learning model is quality assessment data, then the quality of the service provided to the user is determined based on the quality assessment data; if the output of the second machine learning model is recommendation assessment data, then the recommendation assessment data is used to determine whether to recommend a certain item or service to the user.
[0082] It should be noted that, optionally, both the first and second machine learning models in this embodiment can be models obtained through the method described in Embodiment 1 above, and their specific training and acquisition processes will not be repeated here. It should also be noted that the descriptions of the relevant steps in this embodiment are relatively concise, and their specific implementations can also refer to the descriptions of the relevant parts in Embodiment 1 above, and will not be repeated here either.
[0083] The processing procedures of the first and second machine learning models described above in this embodiment are as follows: Figure 3B As shown, by Figure 3B As can be seen, the tabular data corresponding to the user information is processed by the first machine learning model, and the output representation A is output; the output representation A and the user information are input into the second machine learning model; after processing by the second machine learning model, the corresponding service processing data Output is output.
[0084] The following example, using a telecommunications operation scenario, illustrates the above process.
[0085] In this scenario, a user calls the telecommunications operator's customer service hotline. The telecommunications operator's backend server obtains the user's identification information, such as the user ID. Then, based on the user ID, it determines the user's corresponding information, including: user behavior characteristics on a daily basis (e.g., whether they have called the operator's customer service hotline, whether they have purchased a telecommunications package, etc.), user behavior characteristics on a monthly basis (e.g., the user's monthly behavior characteristics over the past two years, such as whether they have visited telecommunications operator locations, the type of telecommunications package they use, etc.), and user call-in behavior characteristics (e.g., whether they have called for consultations, the details of the consultation, etc.). This information is then converted into corresponding tabular data. This tabular data is input into a first machine learning model, which outputs corresponding representation data. Next, this representation data, along with the original user information, is input into a second machine learning model. In this example, the second machine learning model is set up to evaluate and recommend telecommunications packages. Assuming the second machine learning model's evaluation outputs a recommendation probability of 0.8 for a certain telecommunications package X, it will generate corresponding information, such as "It is recommended to recommend telecommunications package X to this user," and feed this information back to the telecommunications operator's customer service. Upon receiving this information, the telecommunications operator's customer service will, based on this, respond to the user's inquiries or questions and recommend telecommunications package X to the user.
[0086] This embodiment can leverage the first and second machine learning models to provide effective service processing, thereby improving service efficiency and user experience.
[0087] Example 3
[0088] Reference Figure 4A The flowchart illustrates the steps of a data processing method according to Embodiment 3 of this application.
[0089] This embodiment uses a customer service scenario as an example to illustrate the data processing method provided in this application embodiment. The data processing method of this embodiment includes the following steps:
[0090] Step S100: Receive customer service inquiries via inbound calls.
[0091] Many businesses and service platforms have customer service hotlines, which users can call for service inquiries or to file complaints. When a customer calls the hotline, the customer service terminal receives the call, thus initiating a customer service consultation call.
[0092] Step S200: Based on the customer service inquiry call, obtain the corresponding call content and customer information, and send the call content and customer information to the server.
[0093] In this embodiment, customer information can be implemented as customer identification information carried during customer service calls. However, it is not limited to this. In some cases, the customer service terminal can also obtain other customer information based on the customer identification, such as customer attribute information, customer call history, etc. In this case, customer information may include the aforementioned other information in addition to the customer identification.
[0094] Since customer service calls are mostly voice calls, the text content corresponding to the voice can be obtained through appropriate means such as speech recognition or ASR as the acquired call content.
[0095] Step S300: Receive the marketing strategy for the customer determined by the receiving server after processing the call content and customer information based on a pre-trained machine learning model.
[0096] On the one hand, customer information can be enriched and supplemented based on the call content; on the other hand, customer information can be updated. For example, user behavior characteristics information of the customer on a daily basis can be updated. However, it is not limited to this; the call content can also be sent to the server for storage only, or the relevant information can be updated later.
[0097] In one feasible approach, this step can be implemented as follows: receiving a marketing strategy for the customer determined by the server, wherein the marketing strategy is obtained by the server based on the customer information, corresponding customer communication feature information, and inputting the call content and customer communication feature information into a pre-trained machine learning model, and obtaining the strategy based on the output of the machine learning model. Optionally, the customer communication feature information input into the machine learning model is information that meets a preset fixed format, such as tabular data. Optionally, the customer communication feature information includes at least one of the following: customer attribute information, customer historical communication consumption information (such as communication package purchase behavior information, data consumption information, call consumption information, etc.), and customer consultation and consumption conversion information (such as the probability or number of times a communication consumption product (such as a communication package) is successfully recommended to the customer through a customer service hotline, etc.).
[0098] In this embodiment, the server-side machine learning model includes the first machine learning model and the second machine learning model trained in the above embodiments. The specific processing of the input data by the first and second machine learning models can be found in the descriptions of the relevant parts in the foregoing embodiments, and will not be repeated here.
[0099] Subsequently, after the client receives the message recommendation information returned by the server, it can make targeted consumption recommendations to customers based on this marketing strategy, thereby improving the efficiency of consultation and consumption conversion.
[0100] The following example, using a telecommunications operation scenario, illustrates the above process. Figure 4B As shown.
[0101] Figure 4BIn this example, customer X dials the customer service hotline using their client device (such as a mobile phone). Upon receiving the call, the customer service provider responds to customer X's call and simultaneously obtains customer X's identifier. This identifier, along with the call data, is then sent to the server. After obtaining the call data and customer X's identifier, the server retrieves customer X's communication characteristics, such as age, city, location, and other attributes; user behavior characteristics on a daily basis (e.g., whether they have called the customer service hotline, purchased a communication package, etc.); user behavior characteristics on a monthly basis (e.g., monthly behavior characteristics over the past two years, such as whether they have visited communication service locations, their communication package usage, etc.); and user call-in behavior characteristics (e.g., whether they have made inquiries, and the details of those inquiries). In this example, the call data received by the server can be stored on the server and, optionally, used to update customer X's communication characteristics later. Next, the server inputs customer X's communication characteristics into the first machine learning model and obtains the corresponding representation data output by the first machine learning model. Then, the server inputs this representation data and customer X's communication characteristics into the second machine learning model to obtain the probability of successful recommendation of at least one communication product. In this example, it is assumed that the probability of successful recommendation of communication products indicates that communication package A has the highest success rate, followed by B. This information will then be fed back to the customer service client by the server. After receiving this information, the customer service representative will recommend it to customer X during the current call. For example, the customer service representative can recommend communication package A to customer X during the call; if unsuccessful, they can try recommending communication package B.
[0102] As can be seen, this embodiment can greatly improve the efficiency of consultation and consumption conversion in customer service scenarios.
[0103] It should be noted that in many embodiments of this application, all information related to users or customers is obtained with authorization and permission. Of course, this is not the only possibility; it can also be obtained through sample simulation.
[0104] Example 4
[0105] Reference Figure 5 The diagram shows a structural schematic of an electronic device according to Embodiment 4 of this application. The specific embodiments of this application do not limit the specific implementation of the electronic device.
[0106] like Figure 5 As shown, the electronic device may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.
[0107] in:
[0108] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408.
[0109] Communication interface 504 is used to communicate with other electronic devices or servers.
[0110] The processor 402 is used to execute program 410, specifically to perform the relevant steps in the above data processing method embodiment.
[0111] Specifically, program 410 may include program code that includes computer operation instructions.
[0112] Processor 402 may be a CPU, an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application. The smart device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.
[0113] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.
[0114] Specifically, program 410 can be used to cause processor 402 to perform operations corresponding to the data processing method described in any of the foregoing method embodiments.
[0115] The specific implementation of each step in program 410 can be found in the corresponding steps and units described in the above data processing method embodiments, and has corresponding beneficial effects, which will not be repeated here. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding process descriptions in the foregoing method embodiments, and will not be repeated here.
[0116] This application also provides a computer program product, including computer instructions that instruct a computing device to perform an operation corresponding to any of the data processing methods in the above-described multiple method embodiments.
[0117] It should be noted that, depending on the implementation needs, the various components / steps described in the embodiments of this application can be broken down into more components / steps, or two or more components / steps or parts of the operation of components / steps can be combined into new components / steps to achieve the purpose of the embodiments of this application.
[0118] The methods described in the embodiments of this application can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as a CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code downloaded over a network that is originally stored in a remote recording medium or a non-transitory machine-readable medium and will be stored in a local recording medium. Thus, the methods described herein can be stored as software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code that, when accessed and executed by the computer, processor, or hardware, implements the data processing methods described herein. Furthermore, when a general-purpose computer accesses code used to implement the data processing methods shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for executing the data processing methods shown herein.
[0119] Those skilled in the art will recognize that the units and method steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the embodiments of this application.
[0120] The above embodiments are only used to illustrate the embodiments of this application, and are not intended to limit the embodiments of this application. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of the embodiments of this application. Therefore, all equivalent technical solutions also fall within the scope of the embodiments of this application, and the patent protection scope of the embodiments of this application should be defined by the claims.
Claims
1. A data processing method, comprising: According to a preset fixed format, the sample data of call content representing users conducting customer service inquiries in communication operation scenarios and the corresponding user information are converted into sample format data. The sample format data is subjected to noise exchange processing to obtain noisy sample format data; A first training sample is constructed based on the noisy sample format data, and a first machine learning model for feature extraction is trained. The first machine learning model is an encoder model based on tabular data, which is configured to perform dimensionality upscaling on the input noisy sample format data to extract features and output sample representation data. The sample representation data output by the first machine learning model after training is used to construct the second training samples for the second machine learning model for evaluating communication package recommendations.
2. The method according to claim 1, wherein, The method further includes: Acquire labeled sample data and sample representation data obtained by the first machine learning model after training it to extract features from the sample format data corresponding to the labeled sample data; A second training sample is constructed based on the labeled sample data, the labels corresponding to the labeled sample data, and the sample representation data corresponding to the labeled sample data, and the second machine learning model used for service processing is trained.
3. The method according to claim 1, wherein, The step of performing noise processing on the sample format data to obtain noisy sample format data includes: The sample format data is subjected to column-swapping noise processing to obtain noisy sample format data.
4. The method according to claim 3, wherein, The step of constructing a first training sample based on the noisy sample format data and training a first machine learning model for feature extraction includes: Construct the first training sample based on the noisy sample format data; The first training sample is input into the input layer of the encoder; The encoder upscales the first training sample received by the input layer through its hidden layer, and then extracts features based on the upscaled first training sample. The encoder's output layer outputs sample representation data corresponding to the first training sample based on the feature extraction result.
5. The method according to any one of claims 1-4, wherein, The sample data includes the following information: user attribute information and / or user behavior information.
6. A data processing method, comprising: Based on the user identifier of the user to be served, determine the user information corresponding to the user identifier, wherein the user to be served is a user who is conducting customer service consultation in a communication operation scenario; The fixed-format data corresponding to the user information is input into a first machine learning model for feature extraction to obtain the representation data output by the first machine learning model. The user information and the representation data are input into a second machine learning model used for communication package recommendation and evaluation to obtain service processing data output by the second machine learning model. Based on the service processing data, provide the user with corresponding service information; Wherein, the first machine learning model is a model obtained according to any one of the data processing methods described in claims 1-5, and the second machine learning model is a model obtained according to any one of the data processing methods described in claims 2-5.
7. A data processing method, comprising: Receive customer service inquiries via inbound calls; Based on the customer service call, obtain the corresponding call content and customer information corresponding to the customer, and send the call content and customer information to the server. The system receives a marketing strategy for the customer determined by the server based on a pre-trained machine learning model after processing the call content and the customer information. The data processing method applies the model obtained by the data processing method as described in any one of claims 1-5.
8. A computer storage medium having a computer program stored thereon, which, when executed by a processor, implements the data processing method as described in any one of claims 1-7.