Information acquisition method and device, computer readable storage medium, and electronic device

Abstract

The present disclosure relates to the field of computers, and provides an information acquisition method, device, computer-readable storage medium, and electronic device, the method comprising: obtaining a multidimensional feature data set corresponding to a service request, wherein the multidimensional feature data set includes multiple feature data of multiple dimensions; obtaining a feature vector set of the multidimensional feature data set, and inputting the feature vector set of the multidimensional feature data set into a category prediction model to obtain a prediction category corresponding to the multidimensional feature data set, and a prediction probability corresponding to the prediction category; obtaining target category information corresponding to the service request according to the prediction category and the prediction probability corresponding to the prediction category. The present disclosure obtains target category information of a service request through feature data of multiple dimensions, thereby improving the accuracy of information acquisition.

Description

Background Art

[0002] With the development of computer technology, many large enterprises, government departments and other institutions use computer technology to submit forms for related issues when dealing with customer issues or cross-departmental collaborative issues. This method can clearly record and save relevant issues and solutions, avoiding the omission of related issues.

[0003] In the prior art, the form is often filled in manually, or the default option with the most number of fills is built into the options to help people choose. When there are many options and the person filling in the form is not familiar with the relevant content, it will take a lot of time and the accuracy rate is low.

[0004] In view of this, there is an urgent need in the art to develop a new information acquisition method and device.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of the present disclosure is to provide an information acquisition method, an information acquisition device, a computer-readable storage medium and an electronic device, thereby improving the accuracy and efficiency of information acquisition at least to a certain extent.

[0007] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0008] According to one aspect of the present disclosure, there is provided an information acquisition method, the method comprising: obtaining a multidimensional feature data set corresponding to a service request, wherein the multidimensional feature data set comprises a plurality of feature data of a plurality of dimensions; obtaining a feature vector set of the multidimensional feature data set, and inputting the feature vector set of the multidimensional feature data set into a category prediction model to obtain a predicted category corresponding to the multidimensional feature data set, and a predicted probability corresponding to the predicted category; and obtaining target category information corresponding to the service request according to the predicted category and the predicted probability corresponding to the predicted category.

[0009] In some exemplary embodiments of the present disclosure, the category prediction model includes a department category prediction model and a business category prediction model, and the prediction category includes a predicted department category and a predicted business category; the feature vector set of the multidimensional feature data set is input into the category prediction model to obtain the predicted category corresponding to the multidimensional feature data set, and the predicted probability corresponding to the predicted category, including: inputting the feature vector set of the multidimensional feature data set into the department category prediction model to obtain the predicted department category corresponding to the multidimensional feature data set, and the first prediction probability corresponding to the predicted department category; inputting the feature vector set of the multidimensional feature data set and the predicted department category into the business category prediction model to obtain the predicted business category corresponding to the multidimensional feature data set, and the second prediction probability corresponding to the predicted business category.

[0010] In some exemplary embodiments of the present disclosure, the predicted department categories include multiple ones; the feature vector set of the multidimensional feature data set and the predicted department categories are input into the business category prediction model, including: sorting the multiple predicted department categories according to the first prediction probability corresponding to the predicted department categories, and obtaining a preset number of the predicted department categories as target predicted department categories; obtaining the data vector of the target predicted department category, fusing the data vector of the target predicted department category and the feature vector set of the multidimensional feature data set to obtain the fused feature vector set, and inputting the fused feature vector set into the business category prediction model.

[0011] In some exemplary embodiments of the present disclosure, the predicted business categories include multiple ones; the target category information corresponding to the service request is obtained according to the predicted categories and the predicted probabilities corresponding to the predicted categories, including: obtaining the target predicted business category according to the second prediction probability corresponding to each of the predicted business categories; judging whether the target predicted business category has an affiliation with the target prediction department category corresponding to the target predicted business category according to a preset affiliation; if so, configuring the target predicted business category and the target prediction department category as target category information.

[0012] In some exemplary embodiments of the present disclosure, obtaining a feature vector set of a multidimensional feature data set includes: inputting the multidimensional feature data set into a vector extraction model to obtain a feature vector set corresponding to the multidimensional feature data set.

[0013] In some exemplary embodiments of the present disclosure, the feature data in the multidimensional feature data set includes discrete feature data, continuous feature data, multi-value discrete feature data, and text feature data; the method also includes: if the feature data is the discrete feature data, the continuous feature data, and the multi-value discrete feature data, then configuring the vector extraction model as a feature outline; if the feature data is the text feature data, then configuring the vector extraction model as a word vector model.

[0014] In some exemplary embodiments of the present disclosure, the multidimensional feature data set is input into a vector extraction model to obtain a feature vector set corresponding to the multidimensional feature data set, including: determining whether the feature data in the multidimensional feature data set is the text feature data; if so, inputting the feature data into the word vector model to obtain a first feature vector corresponding to the feature data; if not, inputting the feature data into the feature outline to obtain a second feature vector corresponding to the feature data; and performing vector concatenation of the first feature vector and the second feature vector to obtain a feature vector set corresponding to the multidimensional feature data set.

[0015] In some exemplary embodiments of the present disclosure, the word vector model includes three convolutional neural network layers, three maximum pooling layers, and one fully connected layer.

[0016] In some exemplary embodiments of the present disclosure, the method further includes: acquiring all text data in the resource conversion platform, configuring the text data into text data samples; and training the word vector model to be trained by using the text data samples to obtain the word vector model.

[0017] In some exemplary embodiments of the present disclosure, the method also includes: obtaining a historical multidimensional feature data set corresponding to historical service requests, and obtaining a historical feature vector set corresponding to the historical multidimensional feature data set according to a vector extraction model; configuring the historical feature vector set as a feature vector set sample, and configuring the target category information corresponding to the historical service request as a category label of the feature vector set sample; training the category prediction model to be trained through the feature vector set samples and the category labels of the feature vector set samples to obtain the category prediction model.

[0018] In some exemplary embodiments of the present disclosure, the category prediction model to be trained is trained using the feature vector set samples and the category labels of the feature vector set samples to obtain the category prediction model, including: inputting the feature vector samples into the category prediction model to be trained to obtain the sample category corresponding to the feature vector samples; determining a loss function based on the sample category and the category label, and obtaining the category prediction model by adjusting the parameters of the category prediction model to be trained until the loss function reaches a minimum.

[0019] In some exemplary embodiments of the present disclosure, the category prediction model to be trained includes a department category prediction model to be trained and a business category prediction model to be trained; the feature vector samples are input into the category prediction model to be trained to obtain sample categories corresponding to the feature vector samples, including: inputting the feature vector set samples into the department category prediction model to be trained to obtain multiple sample department categories corresponding to the feature vector set samples, and a first probability corresponding to each of the sample department categories; determining a target sample department category from the multiple sample department categories according to each of the first probabilities; inputting the feature vector set samples and the target sample department category into the business category prediction model to be trained to obtain multiple sample business categories corresponding to the feature vector set samples, and a second probability corresponding to each of the sample business categories.

[0020] In some exemplary embodiments of the present disclosure, the category labels include department category labels and business category labels; determining a loss function based on the sample categories and the category labels, and adjusting the parameters of the category prediction model to be trained until the loss function reaches a minimum to obtain the category prediction model, including: determining a first loss function based on the sample department categories and the department category labels, and adjusting the parameters of the department category prediction model to be trained until the first loss function reaches a minimum to obtain the department category prediction model; and determining a second loss function based on the sample business categories and the business category labels, and adjusting the parameters of the business category prediction model to be trained until the second loss function reaches a minimum to obtain the business category prediction model.

[0021] In some exemplary embodiments of the present disclosure, the method also includes: sending the target category information corresponding to the service request to the business window of the terminal, and receiving a correction operation for the target category information, wherein the correction operation includes real category information; determining a prediction accuracy based on the target category information and the real category information, and judging whether the prediction accuracy is less than a first accuracy threshold; if so, using the multidimensional feature data set as a training sample, and using the real category information as a label corresponding to the training sample, and training the category prediction model through the training sample and the label corresponding to the training sample to obtain an updated category prediction model.

[0022] In some exemplary embodiments of the present disclosure, the method further includes: configuring a flow control switch, allocating the plurality of multidimensional feature data sets to the category prediction model and the updated category prediction model through the flow control switch; monitoring the prediction accuracy of the updated category prediction model in real time, and when the prediction accuracy of the updated category prediction model is greater than a second accuracy threshold, replacing the category prediction model with the updated category prediction model.

[0023] In some exemplary embodiments of the present disclosure, obtaining a multidimensional feature data set corresponding to a service request includes: calling real-time data interfaces of multiple dimensions, and obtaining multiple feature data corresponding to the service request through each of the real-time data interfaces.

[0024] According to one aspect of the present disclosure, there is provided an information acquisition device, comprising: a data acquisition module, used to acquire a multidimensional feature data set corresponding to a service request, wherein the multidimensional feature data set includes multiple feature data of multiple dimensions; a category prediction model, used to acquire a feature vector set of the multidimensional feature data set, and input the feature vector set of the multidimensional feature data set into the category prediction model to obtain a predicted category corresponding to the multidimensional feature data set, and a predicted probability corresponding to the predicted category; an information acquisition module, used to obtain target category information corresponding to the service request according to the predicted category and the predicted probability corresponding to the predicted category.

[0025] According to one aspect of the present disclosure, a computer-readable medium is provided, on which a computer program is stored. When the program is executed by a processor, the information acquisition method as described in the above embodiments is implemented.

[0026] According to one aspect of the present disclosure, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the information acquisition method as described in the above embodiments.

[0027] It can be seen from the above technical solutions that the information acquisition method and device, computer-readable storage medium, and electronic device in the exemplary embodiments of the present disclosure have at least the following advantages and positive effects:

[0028] The information acquisition method disclosed in the present invention first obtains a multidimensional feature data set corresponding to a service request, wherein the multidimensional feature data set includes multiple feature data of multiple dimensions; then, obtains a feature vector set of the multidimensional feature data set, and inputs the feature vector set of the multidimensional feature data set into a category prediction model to obtain a prediction category corresponding to the multidimensional feature data set, and a prediction probability corresponding to the prediction category; finally, obtains target category information corresponding to the service request based on the prediction category and the prediction probability corresponding to the prediction category. The information acquisition method disclosed in the present invention can obtain feature data of multiple dimensions corresponding to a service request, and obtain target category information corresponding to feature data of multiple dimensions through a category prediction model, thereby solving the multi-classification problem, expanding the model coverage, and improving the efficiency and accuracy of obtaining information.

[0029] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0031] Figure 1 The following is a schematic diagram showing a flow chart of an information acquisition method according to an embodiment of the present disclosure;

[0032] Figure 2 A schematic diagram of a method flow for obtaining a feature vector set according to an embodiment of the present disclosure is shown;

[0033] Figure 3 A schematic diagram of a method flow for obtaining a predicted category according to an embodiment of the present disclosure is shown;

[0034] Figure 4 A schematic diagram of a method flow for obtaining target category information according to an embodiment of the present disclosure is shown;

[0035] Figure 5 A schematic diagram of a method flow for obtaining a category prediction model according to an embodiment of the present disclosure is shown;

[0036] Figure 6A schematic diagram of a method flow for training a to-be-trained category prediction model according to an embodiment of the present disclosure is shown;

[0037] Figure 7 A schematic diagram of a method flow for obtaining a sample category according to an embodiment of the present disclosure is shown;

[0038] Figure 8 A schematic diagram of a method flow chart for obtaining an updated category prediction model according to an embodiment of the present disclosure is shown;

[0039] Fig. 9 The schematic diagram of the architecture of the information acquisition system according to an embodiment of the present disclosure is schematically shown;

[0040] Fig.10 A block diagram of an information acquisition device according to an embodiment of the present disclosure is schematically shown;

[0041] Fig.11 The module diagram of the electronic device according to an embodiment of the present disclosure is schematically shown;

[0042] Fig.12 A schematic diagram of a program product according to an embodiment of the present disclosure is shown schematically. DETAILED DESCRIPTION

[0043] 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 this disclosure will be more comprehensive and complete and will fully convey the concept of the example embodiments to those skilled in the art.

[0044] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. 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 without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the present disclosure.

[0045] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0046] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0047] In one embodiment of the present disclosure, an information acquisition method is proposed. The information acquisition method can be applied to but not limited to the following scenarios: on an e-commerce platform, a user's consultation question generation form is pushed to the corresponding production department; or on an enterprise or government complaint platform, the relevant department is determined according to the user's complaint content, etc. The present disclosure does not specifically limit the specific application scenario of the information acquisition method. Changes in specific application scenarios should be understood as falling within the scope of protection of the present disclosure. The present disclosure embodiment takes the application to an e-commerce platform as an example. Figure 1 A flow chart of the information acquisition method is shown, as shown in Figure 1 As shown, the information acquisition method at least includes the following steps:

[0048] Step S110: obtaining a multidimensional feature data set corresponding to the service request, wherein the multidimensional feature data set includes a plurality of feature data in a plurality of dimensions;

[0049] Step S120: obtaining a feature vector set of the multidimensional feature data set, and inputting the feature vector set of the multidimensional feature data set into a category prediction model to obtain a predicted category corresponding to the multidimensional feature data set and a predicted probability corresponding to the predicted category;

[0050] Step S130: obtaining target category information corresponding to the service request according to the predicted category and the predicted probability corresponding to the predicted category.

[0051] The information acquisition method in the disclosed embodiment can acquire feature data of multiple dimensions corresponding to a service request, and obtain target category information corresponding to the feature data of multiple dimensions through a category prediction model, thereby solving the multi-classification problem, expanding the model coverage, and improving the efficiency and accuracy of obtaining information.

[0052] It should be noted that the information acquisition method of the exemplary embodiment of the present disclosure can be executed by a server, and the corresponding information acquisition device of the information acquisition method can also be configured in the server. In addition, it should be understood that a terminal device (e.g., a mobile phone, a tablet, etc.) can also implement each step of the information acquisition method, and the corresponding device can also be configured in the terminal device.

[0053] In order to make the technical solution of the present disclosure clearer, each step of the information acquisition method is described below.

[0054] In step S110, a multidimensional feature data set corresponding to the service request is obtained, wherein the multidimensional feature data set includes a plurality of feature data in a plurality of dimensions.

[0055] In an exemplary embodiment of the present disclosure, a service request includes a service request triggered by a user at a customer service center of an e-commerce platform, and the service request may be made before, during, or after a sale. For example, after a user completes an order, a service request may be made to a customer service center for the order, and the service request may be for any aspect of the order, which is not specifically limited in the present disclosure.

[0056] If a user asks the customer service center multiple different questions about the same order, multiple service requests will be generated, and each service request corresponds to an order number. It can be seen that the same order or the same product can correspond to one or more service requests.

[0057] In addition, the service request can also be triggered by the customer service personnel of the customer service center. When the user asks a question to the customer service center, if the customer service personnel can give a reasonable answer to the question raised by the user, the service request will not be triggered; if the customer service personnel cannot directly give a reasonable answer to the question raised by the user, and needs to query the database corresponding to the target category information of the question or needs to send the question to the personnel corresponding to the target category information of the question for processing, then the customer service personnel can trigger the service request. Among them, the database can be a distributed file system.

[0058] In an exemplary embodiment of the present disclosure, the multidimensional feature data set includes multiple feature data of multiple dimensions. The multiple dimensions include dimensions related to orders and problems on the e-commerce platform, for example, the multiple dimensions may include order dimensions, after-sales dimensions, invoice dimensions, event dimensions, and logistics dimensions. Among them, the event dimension is based on questions raised by users and the communication process between users and customer service personnel.

[0059] In addition, the characteristic data corresponding to each dimension include: the order dimension includes the sales order type code, item price, order comprehensive status code, order flag, order completion time, order delivery time, whether the order is printed, whether the order is picked up, whether the order is packed, whether the order is shipped, the three items purchased by the user recently, etc.; the after-sales dimension includes the after-sales service order link code, the after-sales service order status code, the review result code, the processing result code, the arbitration status, whether there is a dispute, etc.; the invoice dimension includes the invoice status, invoice type, invoice header type, invoice qualification verification, whether the invoice has been mailed, whether the invoice agency is the same as the sales agency, etc.; the event summary includes the event summary, event type, event status, event source, special problem ID, event summary log, etc.; the logistics dimension includes the waybill type code, warehouse number, user's expected delivery date, outbound event, logistics routing node, whether it is the first sorting and delivery, actual operation outlets, actual operation type, etc.

[0060] In an exemplary embodiment of the present disclosure, real-time data interfaces of multiple dimensions are called, and multiple feature data corresponding to the service request are acquired through each real-time data interface.

[0061] Specifically, a service request is received, and a multi-dimensional feature data set is obtained in real time according to the service request. When users purchase items and ask questions on the e-commerce platform, the server will store the feature data of multiple dimensions of the same order in multiple storage units in real time. If the feature data of a certain dimension changes, the feature data of a certain dimension corresponding to the order will also be updated in real time. For example, after the order is completed, the feature data of the logistics dimension will be updated in real time with the transportation of the items; or in the event dimension, during the communication between the user and the customer service staff, the feature of the event summary will also be updated in real time.

[0062] Therefore, in the online application stage, in response to the service request, the real-time data interface of multiple dimensions is called according to the order number corresponding to the service request, and the multiple feature data of multiple dimensions corresponding to the order number are obtained through the real-time data interface of multiple dimensions respectively, and the multiple feature data of multiple dimensions generate a multidimensional feature data set.

[0063] In addition, a service request corresponds to a work order number, which is a unique identifier of the service request. For an order number, a user or customer service staff can issue a service request at one or more times, and an order number can correspond to multiple service requests. Therefore, the work order number can be composed of the order number and the current timestamp of issuing the service request, or it can be any string that can uniquely identify the service request. The present disclosure does not specifically limit the composition of the work order number.

[0064] In addition, after obtaining the multidimensional feature data set corresponding to the service request, the work order number corresponding to the service request and the corresponding multidimensional feature data set are stored in the database in correspondence, so that the multidimensional feature data set corresponding to the service request can be queried by the work order number, and can be used to train the category prediction model.

[0065] In an exemplary embodiment of the present disclosure, after obtaining a multidimensional feature data set corresponding to a service request, the multidimensional feature data set may be preprocessed. The preprocessing may include: determining whether the feature data corresponding to the event dimension in the multidimensional feature data set is empty, and if the feature data corresponding to the event dimension in the multidimensional feature data set is empty, discarding the multidimensional feature data set. In addition, the preprocessing may also include: completing data whose feature data is empty. The present disclosure does not specifically limit the method of completing the feature data and the method of preprocessing the feature data.

[0066] In step S120, a feature vector set of the multidimensional feature data set is obtained, and the feature vector set of the multidimensional feature data set is input into a category prediction model to obtain a predicted category corresponding to the multidimensional feature data set and a predicted probability corresponding to the predicted category.

[0067] In an exemplary embodiment of the present disclosure, the prediction category refers to the form category corresponding to the service request, and the form category may include multiple department categories, and the form category may also include multiple business categories. In addition, the form category may also include a major category and a minor category, each major category includes multiple minor categories, the major category refers to the department category, and each department category may include multiple business categories.

[0068] For example, in an e-commerce platform, department categories may include logistics and distribution, spare parts warehouse, warehousing, procurement and sales, large household appliances, bills, and so on. The logistics and distribution category may include business categories such as customer urging orders, damaged goods, consultation and self-pickup; the spare parts warehouse category may include: missing parts, wrong parts, damaged parts, order cancellation, wrong goods for the entire order, and other business categories; the warehousing category may include: damaged, missing commodity B, urging orders A, and other business categories; the procurement and sales category may include: factory direct delivery order problems, after-sales three guarantees and repairs, arrival consultation, and other business categories; the large household appliances category may include: forward urging order problems, unpacking and inspection, reverse urging order problems, and other business categories; the bill category may include: unsuccessful invoice self-service order entry, electronic invoices, historical general invoices, and other business categories. Of course, specific department categories and business categories can be set according to the application scenario, and this disclosure does not make specific limitations on this.

[0069] In an exemplary embodiment of the present disclosure, the feature data in the multidimensional feature data set includes discrete feature data, continuous feature data, multi-value discrete feature data, and text feature data, wherein discrete feature data, continuous feature data, multi-value discrete feature data, and text feature data are data formats of feature data.

[0070] For example, among the feature data of multiple dimensions listed in the above embodiment, the audit result code of the after-sales dimension belongs to discrete feature data, the three items recently purchased by the user in the order dimension belong to multi-value discrete feature data, the item price in the order dimension belongs to continuous feature data, the event summary in the event dimension belongs to text feature data, etc. The embodiment of the present disclosure lists an example of discrete feature data, continuous feature data, multi-value discrete feature data, and text feature data respectively. Of course, among the multiple feature data, there are also multiple feature data of discrete feature data, continuous feature data, multi-value discrete feature data, and text feature data, which are not specifically limited in the present disclosure.

[0071] In addition, since the feature data of multiple dimensions include multiple data formats, it is necessary to convert the feature data of multiple dimensions into a specific format, and different data conversion methods are used for different data formats. Among them, the specific format includes converting each data format into a vector form.

[0072] In an exemplary embodiment of the present disclosure, a multidimensional feature data set is input into a vector extraction model to obtain a feature vector set corresponding to the multidimensional feature data set. Among them, since the feature data in the multidimensional feature data set includes discrete feature data, continuous feature data, multi-value discrete feature data, and text feature data, the vector extraction model can be configured according to the different data formats of the feature data. If the feature data is discrete feature data, continuous feature data, and multi-value discrete feature data, the vector extraction model is configured as a feature outline; if the feature data is text feature data, the vector extraction model is configured as a word vector model.

[0073] Specifically, Figure 2 Schematic diagram of the method flow for obtaining a feature vector set is shown in FIG. Figure 2 As shown, the process at least includes steps S210 to S240, which are described in detail as follows:

[0074] In step S210, it is determined whether the feature data in the multidimensional feature data set is text feature data.

[0075] In an exemplary embodiment of the present disclosure, it may be determined whether a plurality of feature data in the multi-dimensional feature data set is text feature data according to the data format of the feature data.

[0076] In step S220, if the feature data is text feature data, the feature data is input into a word vector model to obtain a first feature vector corresponding to the feature data.

[0077] In an exemplary embodiment of the present disclosure, the feature outline may be used to configure the vector dimension of the text feature data. After the text feature data is processed using the feature outline, it is input into a word vector model to obtain a feature vector corresponding to the text feature data.

[0078] In an exemplary embodiment of the present disclosure, the word vector model may be a word2vec (word to vector) model. The word2vec model can be trained using the Baidu Encyclopedia dataset, based on the rich corpus in Baidu Encyclopedia, but for specific business scenarios, most corpus have nothing to do with the business scenario, making the word vector model to be trained too large, and loading the model when calling the word vector model to be trained is time-consuming, and the trained word vector model cannot better focus on the business scenario.

[0079] Therefore, in an exemplary embodiment of the present disclosure, when training the word vector model to be trained, all text data in the resource conversion platform are obtained, and the text data are configured as text data samples; the word vector model to be trained is trained with the text data samples to obtain a word vector model.

[0080] The resource conversion platform includes an e-commerce platform, and all text data in the e-commerce platform are obtained. All text data can cover all texts in the business scenario, such as user search texts on the e-commerce platform, user evaluation texts for items or merchants, description texts for items when merchants register, rule texts of the e-commerce platform, text feature data in historical multidimensional feature data sets, etc., which are not specifically limited in this disclosure.

[0081] In an exemplary embodiment of the present disclosure, the word vector model can be composed of three convolutional neural network layers, three maximum pooling layers, and one fully connected layer, and a dropout operation can also be added to the word vector model. The word vector model in this embodiment is used to process text feature data, and the feature vector corresponding to the text feature data is obtained after processing through a multi-layer network structure. This embodiment deletes useless data to a large extent, ensures that the amount of data of the trained word vector model is small, and the time required to call the word vector model is shorter. At the same time, it also makes the word vector model more focused on the text feature data in the business, and the feature vector corresponding to the obtained text feature data is more accurate.

[0082] In step S230, if the feature data is not text feature data, the feature data is input into the feature outline to obtain a second feature vector corresponding to the feature data.

[0083] In an exemplary embodiment of the present disclosure, in order to facilitate format conversion of feature data in different data formats, after a multidimensional feature data set is acquired, the feature data in the multidimensional feature data set is first converted using a feature outline.

[0084] Specifically, data mapping is performed for discrete feature data. For example, the audit results of the after-sales dimension are mapped to data. If the audit is passed, it corresponds to 0; if it is under review, it corresponds to 1; if the audit is not passed, it corresponds to 2.

[0085] For multi-valued discrete data, the corresponding rules are the same as those for discrete data. For example, in the order dimension, a user recently purchased three items, item A corresponds to 0, item B corresponds to 1, and item C corresponds to 2.

[0086] For continuous data, save the average value of all values ​​under the feature data. For example, for the item price in the order dimension, the average of multiple consecutive item prices is 115.6.

[0087] In addition, the feature outline can also configure the vector dimension size of the text feature data. For example, the vector dimension size of the event summary in the event dimension belonging to the text feature data is configured to be 300 dimensions.

[0088] After the multidimensional feature data set in this embodiment is processed by the feature outline, the discrete feature data will save the number of different values ​​​​taken by each feature data and the one-to-one correspondence between the feature data and the feature vector; the multi-valued discrete data and text feature data are constructed in the same structure, and both need to save the size of the vocabulary, the maximum text length, and the vector dimension; the continuous feature data will save the mean of each continuous feature data, as well as the corresponding relationship.

[0089] In step S240, the first feature vector and the second feature vector are concatenated to obtain a feature vector set corresponding to the multi-dimensional feature data set.

[0090] In an exemplary embodiment of the present disclosure, a plurality of first feature vectors and second feature vectors corresponding to a plurality of feature data in a multidimensional feature data set are vector-concatenated to obtain a feature vector set corresponding to the multidimensional feature data set.

[0091] In an exemplary embodiment of the present disclosure, the category prediction model includes a department category prediction model and a business category prediction model. Among them, the department category prediction model and the business category prediction model can be an extremely deep factorization machine model (xDeepFM). The xDeepFM model draws on the design of models such as Wide&Deep and DeepFM, and combines CIN with a linear regression unit and a fully connected neural network unit. The xDeepFM model has the memory capacity of the offline model and inherits the explicit intersection of the multi-dimensional feature data set. It also takes into account the implicit feature intersection and generalization capabilities of the DNN network. Pooling calculations are also used in the CIN network for dimensionality reduction, which effectively avoids the situation of dimensionality explosion.

[0092] In an exemplary embodiment of the present disclosure, the forecast category includes a forecast department category and a forecast business category. Figure 3 A schematic diagram of a method flow chart for obtaining a predicted category is shown, and the process at least includes steps S310 to S320, which are described in detail as follows:

[0093] In step S310, the feature vector set of the multidimensional feature data set is input into the department category prediction model to obtain the predicted department category corresponding to the multidimensional feature data set and the first prediction probability corresponding to the predicted department category.

[0094] In an exemplary embodiment of the present disclosure, the first predicted probability represents the probability that the business department category corresponding to the multidimensional feature data set is a real department category. The larger the first predicted probability is, the greater the probability that the predicted department category is a real department category.

[0095] In an exemplary embodiment of the present disclosure, a department category prediction model is used to obtain multiple predicted department categories corresponding to a multidimensional feature data set. When the multidimensional feature data set and the predicted department categories are input into the business category prediction model, all predicted department categories can be input into the business category prediction model to obtain multiple predicted business categories corresponding to each predicted department category.

[0096] In addition, the forecast department categories may be screened according to the first forecast probability corresponding to each forecast department category, and the forecast department category with a higher first forecast probability may be used as the target forecast department category, and the target forecast department category may be input into the business category forecasting model.

[0097] Specifically, first, a plurality of prediction department categories are sorted according to the first prediction probabilities corresponding to the prediction department categories, and a preset number of prediction department categories are obtained as target prediction department categories.

[0098] Among them, multiple prediction department categories are sorted according to the size of the first prediction probability, and the first prediction probability corresponding to the prediction department category in the front is high, and the first prediction probability corresponding to the prediction department category in the back is low. In addition, the preset number can be set according to the actual situation. For example, the preset number can be set to 1, that is, the prediction department category with the highest first prediction probability is selected as the target prediction department category; the preset number can also be set to 3, that is, the prediction department category with the first prediction probability in TOP (3) is selected as the target prediction department category. The larger the value of the preset number, the more target prediction department categories are obtained, and the larger the subsequent calculation amount is. The present disclosure does not make specific limitations on this.

[0099] Then, the data vector of the target prediction department category is obtained, and the data vector of the target prediction department category and the feature vector set of the multidimensional feature data set are fused to obtain a fused feature vector set, and the fused feature vector set is input into the business category prediction model. Among them, the target prediction department category is vectorized to obtain the data vector of the target prediction department category.

[0100] In step S320, the feature vector set of the multidimensional feature data set and the predicted department category are input into the business category prediction model to obtain the predicted business category corresponding to the multidimensional feature data set and the second prediction probability corresponding to the predicted business category.

[0101] In an exemplary embodiment of the present disclosure, a feature vector set of a multidimensional feature data set is fused with a data vector of a predicted department category to obtain a fused feature vector set, and the fused feature vector set is input into a business category prediction model.

[0102] In an exemplary embodiment of the present disclosure, the data vector of the target prediction department category and the feature vector set of the multidimensional feature data set are fused to obtain a fused feature vector set, and the fused feature vector set is input into the business category prediction model. Specifically, a preset number of target prediction department categories and multidimensional feature data sets are fused to obtain a preset number of fused feature vector sets, and the preset number of fused feature vector sets are input into the business category prediction model to obtain multiple predicted business categories corresponding to each fused feature vector set, and a second prediction probability corresponding to each predicted business category.

[0103] In an exemplary embodiment of the present disclosure, the second prediction probability represents the probability that the predicted business category corresponding to the multidimensional feature data set is a real business category. The larger the second prediction probability is, the greater the probability that the predicted business category is a real business category.

[0104] In step S130, target category information corresponding to the service request is obtained according to the predicted category and the predicted probability corresponding to the predicted category.

[0105] In an exemplary embodiment of the present disclosure, the prediction category is a prediction department category or a prediction business category. It is determined whether the prediction probability corresponding to the prediction category is greater than a prediction probability threshold. If the prediction probability is greater than the prediction probability threshold, the prediction category is used as the target category information corresponding to the service request.

[0106] Specifically, if the prediction probabilities corresponding to multiple prediction categories are greater than the prediction probability threshold, the service request corresponds to multiple target category information, which indicates that the service request is related to multiple department categories or multiple business categories.

[0107] In an exemplary embodiment of the present disclosure, the forecast category includes a forecast department category and a forecast business category. Figure 4 A schematic diagram of a method flow for obtaining target category information is shown, and the flow at least includes steps S410 to S430, which are described in detail as follows:

[0108] In step S410, a target predicted business category is obtained according to the second predicted probability corresponding to each predicted business category.

[0109] In an exemplary embodiment of the present disclosure, multiple predicted business categories are sorted according to the second predicted probabilities corresponding to the predicted business categories, and the predicted business category with the second predicted probability at TOP(n) is obtained as the target predicted business category. The value of n can be set according to actual conditions. The larger the value of n, the more target predicted business categories are obtained, and the larger the subsequent calculation amount is. The present disclosure does not make specific limitations on this.

[0110] In step S420, it is determined whether the target forecast business category and the target forecast department category corresponding to the target forecast business category have a belonging relationship according to the preset belonging relationship.

[0111] In the exemplary embodiment of the present disclosure, since each department category includes multiple business categories, the department category and the business category have an affiliation relationship, and the affiliation relationship between the department category and the business category can be preset, and the affiliation relationship is stored in a database in a distributed format. For example, the department category and the business category are stored in the form of a key-value pair, with the department category as the key value and the business category as the value value.

[0112] Specifically, determining whether a target prediction business category has an affiliation with a target prediction department category corresponding to the target prediction business category may include: obtaining all value values ​​of the target prediction department category, and matching the target prediction business category with all value values ​​of the target prediction department category respectively; if the target prediction business category matches the value value of the target prediction department category, then determining that the target prediction business category has an affiliation with the target prediction department category; if the target prediction business category does not match the value value of the target prediction department category, then determining that the target prediction business category has no affiliation with the target prediction department category.

[0113] In step S430, if the target forecast business category and the target forecast department category have an affiliation relationship, the target forecast business category and the target forecast department category are configured as target category information.

[0114] In an exemplary embodiment of the present disclosure, the target category information corresponding to the service request includes a target forecast business category and a target forecast department category, which may include one or more target forecast department categories and target forecast business categories. The number of target forecast department categories and target forecast business categories can be set according to actual conditions. For example, for a more complex service request, it may not be solved by one department or business, and may correspond to multiple departments or businesses. Therefore, the number of target forecast department categories and target forecast business categories can be set to multiple. The service request and the target category information corresponding to the service request are stored in a database.

[0115] In an exemplary embodiment of the present disclosure, Figure 5 A schematic diagram of a method flow for obtaining a category prediction model is shown, and the process includes at least steps S510 to S530, which are described in detail as follows:

[0116] In step S510, a historical multidimensional feature data set corresponding to the historical service request is obtained, and a historical feature vector set corresponding to the historical multidimensional feature data set is obtained according to a vector extraction model.

[0117] In an exemplary embodiment of the present disclosure, according to the work order number corresponding to the historical service request, a historical multi-dimensional feature data set corresponding to the work order number is obtained in a database.

[0118] In step S520, the historical feature vector set is configured as a feature vector set sample, and the target category information corresponding to the historical service request is configured as a category label of the feature vector set sample.

[0119] In an exemplary embodiment of the present disclosure, target category information corresponding to a historical multidimensional feature dataset is obtained in a database, and the target category information is configured as a category label corresponding to a feature vector set of the multidimensional feature dataset.

[0120] In step S530, the to-be-trained category prediction model is trained using the feature vector set samples and the category labels of the feature vector set samples to obtain a category prediction model.

[0121] In an exemplary embodiment of the present disclosure, Figure 6 A schematic diagram of the method flow for training a training category prediction model is shown, such as Figure 6 As shown, the process at least includes steps S610 to S620, which are described in detail as follows:

[0122] In step S610, the feature vector sample is input into the category prediction model to be trained to obtain the sample category corresponding to the feature vector sample.

[0123] In an exemplary embodiment of the present disclosure, the category prediction model to be trained includes a department category prediction model to be trained and a business category prediction model to be trained. The feature vector sample is input into the department category prediction model to be trained to obtain the sample department category corresponding to the feature vector sample; the sample department category and the feature vector sample are input into the business category prediction model to be trained to obtain the sample business category corresponding to the feature vector sample.

[0124] In an exemplary embodiment of the present disclosure, Figure 7 A schematic diagram of the method flow for obtaining sample categories is shown, such as Figure 7 As shown, the process at least includes steps S710 to S730, which are described in detail as follows:

[0125] In step S710, the feature vector set samples are input into the department category prediction model to be trained to obtain a plurality of sample department categories corresponding to the feature vector set samples and a first probability corresponding to each sample department category.

[0126] In step S720 , a target sample department category is determined from among a plurality of sample department categories according to each first probability.

[0127] In an exemplary embodiment of the present disclosure, the sample department categories are sorted according to the first probabilities corresponding to the multiple sample department categories, and the first m sample department categories are taken as the target sample department categories. The value of m can be set according to actual conditions, and the present disclosure does not specifically limit this.

[0128] In step S730, the feature vector set samples and the target sample department category are input into the business category prediction model to be trained to obtain a plurality of sample business categories corresponding to the feature vector set samples and a second probability corresponding to each sample business category.

[0129] In an exemplary embodiment of the present disclosure, feature vector samples of the target sample department category are obtained, feature fusion is performed on the feature vector set samples and the feature vector samples of the target sample department category to obtain fused feature vector set samples, and the fused feature vector set samples are input into the business category prediction model to be trained.

[0130] In step S620, a loss function is determined according to the sample category and the category label, and the category prediction model is obtained by adjusting the parameters of the category prediction model to be trained until the loss function reaches a minimum.

[0131] In an exemplary embodiment of the present disclosure, category labels include department category labels and business category labels, and a department category prediction model is determined by sample department categories and department category labels, and a business category prediction model is determined by sample business categories and business category labels.

[0132] Specifically, a first loss function is determined according to the sample department category and the department category label, and the department category prediction model is obtained by adjusting the parameters of the department category prediction model to be trained until the first loss function reaches a minimum.

[0133] And determine the second loss function according to the sample business category and the business category label, and obtain the business category prediction model by adjusting the parameters of the business category prediction model to be trained until the second loss function reaches the minimum.

[0134] The first loss function and the second loss function are both distillation loss functions. The formula of the distillation loss function is shown in formula (1):

[0135] L=(1-α)H+αD (1)

[0136] Among them, L is the distillation loss function, α is the parameter, H is the cross entropy loss function, and D is the KL divergence. The specific formula is as follows:

[0137]

[0138]

[0139] Among them, P is the category label of the sample, q is the predicted category of the sample, x is the sample identifier, n represents the nth sample, and T is the parameter.

[0140] In the previous training process, the cross entropy loss function was used, and only the difference between the predicted category with a higher probability and the true category was focused on. The similarity was completely constructed through a sufficient number of samples. However, the contribution of small probability results to the loss function is negligible. Therefore, the loss function of this embodiment uses the distillation loss function instead of the ordinary cross entropy loss function. When calculating the loss function, the loss value corresponding to the probability value of other classes is amplified. The distillation loss function introduces the parameter T. By adjusting the size of T, the predicted categories of different classes are considered equally.

[0141] In an exemplary embodiment of the present disclosure, the category prediction model is deployed to the server via the HTTP method of TensorFlow Serving. In addition, in order to ensure that the input data format of the category prediction model is correct, after obtaining the multidimensional feature data set, the feature vector set corresponding to the multidimensional feature data set is first obtained according to the vector extraction model. That is, the feature outline and word vector model are stored in the redis cache, and the feature vector set of the multidimensional feature data set can be quickly obtained to ensure the timeliness of data request and return.

[0142] In the exemplary embodiment of the present disclosure, due to the large number of department categories and business categories, there may be an increase or decrease in department categories and business categories during online operation. Therefore, there will be a misalignment between offline training data and online real-time data, which can be specifically manifested in that the offline training accuracy is higher than the online accuracy. In order to ensure the accuracy of online category prediction, it is necessary to monitor the accuracy of the category prediction model in real time after the category prediction model is online, and update the category prediction model.

[0143] Figure 8 A schematic diagram of a method flow chart for obtaining an updated category prediction model is shown, and the process at least includes steps S810 to S830, which are described in detail as follows:

[0144] In step S810, target category information corresponding to the service request is sent to a service window of the terminal, and a correction operation for the target category information is received, wherein the correction operation includes real category information.

[0145] In an exemplary embodiment of the present disclosure, the terminal refers to the terminal device where the customer service personnel are located, and the predicted target category information is displayed in the business window so that the customer service personnel can review the target category information. If the review is passed, the customer service personnel can click the confirmation operation on the business window, and after the server receives the confirmation operation, it configures the target category information as the real category information; if the review is not passed, the customer service personnel can perform a correction operation on the business window, and the correction operation can be to enter the real category information, or to modify the target category information, and the server receives the real category information corresponding to the correction operation. In addition, you can add a buried point to save the real category information and the target category information, as well as the multi-dimensional feature data set obtained by the real-time data interface, to improve the calculation efficiency of the prediction accuracy.

[0146] In step S820, the prediction accuracy is determined according to the target category information and the actual category information, and it is determined whether the prediction accuracy is less than a first accuracy threshold.

[0147] In an exemplary embodiment of the present disclosure, the total number of service requests issued within a preset time period is counted, as well as the number of target category information corresponding to the service requests that is identical to the actual category information. The prediction accuracy is obtained by dividing the identical number by the total number.

[0148] The preset time period can be set according to the actual situation, for example, it can be set to one minute or five minutes, and the present disclosure does not make any specific restrictions on this. In addition, the first accuracy rate threshold can also be set according to the actual situation, for example, it can be set to 0.9 or 0.95, and the present disclosure does not make any specific restrictions on this.

[0149] In step S830, if the prediction accuracy is less than the first accuracy threshold, the multidimensional feature data set is used as a training sample, and the true category information is used as the label corresponding to the training sample. The category prediction model is trained by the training sample and the label corresponding to the training sample to obtain an updated category prediction model.

[0150] In an exemplary embodiment of the present disclosure, if the prediction accuracy is less than the first accuracy threshold, the multidimensional feature data set obtained by the real-time data interface is used as a training sample to perform online training on the category prediction model. The process of online training of the category prediction model is the same as the training process of the category prediction model to be trained in the above embodiment, and will not be repeated here.

[0151] In an exemplary embodiment of the present disclosure, after obtaining the updated category prediction model, the category prediction model needs to be replaced. However, in order to ensure the stability of the updated category prediction model, it is necessary to first go online for verification and then completely replace the original category prediction model.

[0152] Specifically, a flow control switch is configured, and multiple multidimensional feature data sets are distributed to the category prediction model and the updated category prediction model through the flow control switch; the prediction accuracy of the updated category prediction model is monitored in real time, and when the prediction accuracy of the updated category prediction model is greater than a second accuracy threshold, the updated category prediction model is used to replace the category prediction model.

[0153] The flow control switch can be a random function that can generate a random number between 0 and 1, use the random number as the flow ratio of the update category prediction model, allocate the online flow to the update category prediction model according to the flow ratio, and monitor the prediction accuracy of the update category prediction model in real time. The second accuracy threshold can be set according to actual conditions, for example, it can be set to 0.9 or 0.95, and the present disclosure does not make specific limitations on this.

[0154] In addition, both the first accuracy threshold and the second accuracy threshold correspond to the prediction accuracy of the category prediction model, except that the first accuracy threshold and the second accuracy threshold are applied to different scenarios of this embodiment. The first accuracy threshold is used to determine whether it is necessary to perform online training on the category prediction model that is running online. If the prediction accuracy of the category prediction model is less than the first accuracy threshold, the category prediction model is trained online to obtain an updated category prediction model. The flow control switch is used to control the flow of the category prediction model and the updated category prediction model. The second accuracy threshold is used to determine whether all the flow needs to be diverted to the updated category prediction model. If the prediction accuracy of the updated category prediction model is greater than the second accuracy threshold, the category prediction model is replaced with the updated category prediction model, that is, all the flow is diverted to the updated category prediction model. Of course, the first accuracy threshold and the second accuracy threshold can be the same or different. For example, the first accuracy threshold is 90% and the second accuracy threshold is 98%. When the accuracy of the category prediction model is less than 90%, online training is performed, and category prediction is performed using the category prediction model and the updated category prediction model obtained through online training respectively. When the accuracy of the updated category prediction model is greater than 98%, the category prediction model is replaced by the new updated category prediction model.

[0155] The method of this embodiment avoids the lag in the prediction accuracy of the statistical category prediction model, greatly reduces the instability factors of the online model, makes the online model switching smoother, and better ensures the stable operation of the online model without the user being aware of it.

[0156] In an exemplary embodiment of the present disclosure, the information acquisition method is applied to an information acquisition system. Fig. 9 A schematic diagram of the architecture of the information acquisition system is shown in FIG. Fig. 9 As shown, the information acquisition system includes a model side, a data side and a business side.

[0157] On the model side, first, obtain the historical multidimensional feature data set from the distributed file system, preprocess the historical multidimensional feature data set, and delete useless data;

[0158] Then, the model training is performed, including: obtaining text data on the e-commerce platform, and using the text data to train the word vector model to be trained to obtain the word vector model; and using the feature outline to obtain the feature vector set samples corresponding to the historical multidimensional feature data set and the word vector model, and using the feature vector set samples and the target category information of the historical multidimensional feature data set to train the category prediction model to be trained to obtain the category prediction model;

[0159] Finally, the category prediction model is deployed in the server, and the feature outline and word vector model are stored in the redis cache.

[0160] On the data side, first, the real-time data interface is called to obtain the multidimensional feature data set corresponding to the service request, and the multidimensional feature data set is stored in the distributed file system;

[0161] Next, the vector extraction model in the cache is called, wherein the vector extraction model includes a feature outline and a word vector model, and the multidimensional feature data set is converted using the vector extraction model to obtain a feature vector set of the multidimensional feature data set;

[0162] Finally, the feature vector set is input into the category prediction model to obtain the predicted category and the predicted probability corresponding to the predicted category, and the target category information of the service request is obtained based on the predicted category and the predicted probability.

[0163] On the business side, the target category information is sent to the business window corresponding to the service request, the real category information corresponding to the service request is received, and a tracking point is added to the business side to save the target category information, the real category information, and the multi-dimensional feature data set.

[0164] In addition, the entire information acquisition system is monitored in real time, and the prediction accuracy of the category prediction model is calculated. When the prediction accuracy is lower than the prediction accuracy threshold, the category prediction model is trained online to obtain an updated category prediction model, and the flow control switch is used to update the category prediction model using the updated category prediction model.

[0165] Those skilled in the art will appreciate that all or part of the steps for implementing the above embodiments are implemented as a computer program executed by a CPU. When the computer program is executed by the CPU, the above functions defined by the above method provided by the present invention are performed. The program may be stored in a computer-readable storage medium, which may be a read-only memory, a disk, or an optical disk.

[0166] In addition, it should be noted that the above-mentioned figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0167] The following describes an embodiment of the device of the present disclosure, which can be used to execute the above-mentioned information acquisition method of the present disclosure. For details not disclosed in the embodiment of the device of the present disclosure, please refer to the embodiment of the above-mentioned information acquisition method of the present disclosure.

[0168] Fig.10 A block diagram of an information acquisition device according to an embodiment of the present disclosure is schematically shown.

[0169] Reference Fig.10 As shown, according to an information acquisition device 1000 of an embodiment of the present disclosure, the information acquisition device 1000 includes: a data acquisition module 1001, a category prediction module 1002 and an information acquisition module 1003. Specifically:

[0170] The data acquisition module 1001 is used to acquire a multi-dimensional feature data set corresponding to the service request, wherein the multi-dimensional feature data set includes a plurality of feature data in a plurality of dimensions;

[0171] The category prediction module 1002 is used to obtain a feature vector set of the multidimensional feature data set, and input the feature vector set of the multidimensional feature data set into the category prediction model to obtain a predicted category corresponding to the multidimensional feature data set and a predicted probability corresponding to the predicted category;

[0172] The information acquisition module 1003 is used to obtain target category information corresponding to the service request according to the predicted category and the predicted probability corresponding to the predicted category.

[0173] In an exemplary embodiment of the present disclosure, the data acquisition module 1001 may also be used to call real-time data interfaces of multiple dimensions, and acquire multiple feature data corresponding to the service request through each real-time data interface.

[0174] In an exemplary embodiment of the present disclosure, the category prediction model includes a department category prediction model and a business category prediction model, and the prediction category includes a predicted department category and a predicted business category. The category prediction module 1002 can also be used to input the feature vector set of the multidimensional feature data set into the department category prediction model to obtain the predicted department category corresponding to the multidimensional feature data set, and the first prediction probability corresponding to the predicted department category; input the feature vector set and the predicted department category of the multidimensional feature data set into the business category prediction model to obtain the predicted business category corresponding to the multidimensional feature data set, and the second prediction probability corresponding to the predicted business category.

[0175] In an exemplary embodiment of the present disclosure, the predicted department category includes multiple. The category prediction module 1002 can also be used to sort the multiple predicted department categories according to the first prediction probability corresponding to the predicted department category, and obtain a preset number of predicted department categories as target predicted department categories; obtain the data vector of the target predicted department category, fuse the data vector of the target predicted department category with the feature vector set of the multidimensional feature data set to obtain a fused feature vector set, and input the fused feature vector set into the business category prediction model.

[0176] In an exemplary embodiment of the present disclosure, the forecast business category includes multiple categories. The category forecast module 1002 can also be used to obtain a target forecast business category according to the second forecast probability corresponding to each forecast business category; determine whether the target forecast business category has a relationship with the target forecast department category corresponding to the target forecast business category according to a preset relationship; if so, configure the target forecast business category and the target forecast department category as target category information.

[0177] In an exemplary embodiment of the present disclosure, the category prediction module 1002 may also be used to input the multidimensional feature data set into a vector extraction model to obtain a feature vector set corresponding to the multidimensional feature data set.

[0178] In an exemplary embodiment of the present disclosure, the feature data in the multidimensional feature data set includes discrete feature data, continuous feature data, multi-value discrete feature data, and text feature data. The category prediction module 1002 can also be used to configure the vector extraction model as a feature outline if the feature data is discrete feature data, continuous feature data, and multi-value discrete feature data; if the feature data is text feature data, the vector extraction model is configured as a word vector model.

[0179] In an exemplary embodiment of the present disclosure, the category prediction module 1002 can also be used to determine whether the feature data in the multidimensional feature data set is text feature data; if so, the feature data is input into the word vector model to obtain a first feature vector corresponding to the feature data; if not, the feature data is input into the feature outline to obtain a second feature vector corresponding to the feature data; the first feature vector and the second feature vector are vector-concatenated to obtain a feature vector set corresponding to the multidimensional feature data set.

[0180] In an exemplary embodiment of the present disclosure, the category prediction module 1002 can also be used for a word vector model including three convolutional neural network layers, three maximum pooling layers, and one fully connected layer.

[0181] In an exemplary embodiment of the present disclosure, the information acquisition device 1000 also includes a word vector model training module (not shown in the figure), which is used to acquire all text data in the resource conversion platform and configure the text data into text data samples; the word vector model to be trained is trained through the text data samples to obtain a word vector model.

[0182] In an exemplary embodiment of the present disclosure, the information acquisition device 1000 also includes a category prediction model training module (not shown in the figure), which is used to obtain a historical multidimensional feature data set corresponding to a historical service request, and obtain a historical feature vector set corresponding to the historical multidimensional feature data set according to a vector extraction model; configure the historical feature vector set as a feature vector set sample, and configure the target category information corresponding to the historical service request as the category label of the feature vector set sample; train the category prediction model to be trained using the feature vector set samples and the category labels of the feature vector set samples to obtain a category prediction model.

[0183] In an exemplary embodiment of the present disclosure, the category prediction model training module can also be used to input feature vector samples into the category prediction model to be trained to obtain the sample category corresponding to the feature vector sample; determine the loss function according to the sample category and category label, and obtain the category prediction model by adjusting the parameters of the category prediction model to be trained until the loss function reaches the minimum.

[0184] In an exemplary embodiment of the present disclosure, the category prediction model to be trained includes a department category prediction model to be trained and a business category prediction model to be trained. The category prediction model training module can also be used to input the feature vector set samples into the department category prediction model to be trained to obtain multiple sample department categories corresponding to the feature vector set samples, and a first probability corresponding to each sample department category; determine a target sample department category from the multiple sample department categories according to each first probability; input the feature vector set samples and the target sample department category into the business category prediction model to be trained to obtain multiple sample business categories corresponding to the feature vector set samples, and a second probability corresponding to each sample business category.

[0185] In an exemplary embodiment of the present disclosure, the category label includes a department category label and a business category label. The category prediction model training module can also be used to determine a first loss function based on sample department categories and department category labels, and to obtain a department category prediction model by adjusting the parameters of the department category prediction model to be trained until the first loss function reaches a minimum; and to determine a second rectification loss function based on sample business categories and business category labels, and to obtain a business category prediction model by adjusting the parameters of the business category prediction model to be trained until the second rectification loss function reaches a minimum.

[0186] In an exemplary embodiment of the present disclosure, the information acquisition device 1000 also includes a category prediction model update module (not shown in the figure), which is used to send the target category information corresponding to the service request to the business window of the terminal, and receive a correction operation for the target category information, wherein the correction operation includes a real category information; determine the prediction accuracy based on the target category information and the real category information, and judge whether the prediction accuracy is less than the first accuracy threshold; if so, use the multidimensional feature data set as a training sample, and use the real category information as the label corresponding to the training sample, and train the category prediction model through the training sample and the label corresponding to the training sample to obtain an updated category prediction model.

[0187] In an exemplary embodiment of the present disclosure, the information acquisition device 1000 also includes an online model switching module (not shown in the figure), which is used to configure a flow control switch, and allocate multiple multidimensional feature data sets to the category prediction model and the updated category prediction model through the flow control switch; monitor the prediction accuracy of the updated category prediction model in real time, and when the prediction accuracy of the updated category prediction model is greater than the second accuracy threshold, use the updated category prediction model to replace the category prediction model.

[0188] The specific details of each of the above-mentioned information acquisition devices have been described in detail in the corresponding information acquisition method, so they will not be repeated here.

[0189] It should be noted that, although several modules or units of the device for execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied 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 embodied.

[0190] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0191] It will be appreciated by those skilled in the art that various aspects of the present invention may be implemented as a system, method or program product. Therefore, various aspects of the present invention may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as a "circuit", "module" or "system".

[0192] Refer to the following Fig.11 The electronic device 1100 according to this embodiment of the present invention is described. Fig.11The electronic device 1100 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0193] like Fig.11 As shown, the electronic device 1100 is in the form of a general computing device. The components of the electronic device 1100 may include, but are not limited to: the at least one processing unit 1110, the at least one storage unit 1120, a bus 1130 connecting different system components (including the storage unit 1120 and the processing unit 1110), and a display unit 1140.

[0194] The storage unit stores program codes, which can be executed by the processing unit 1110, so that the processing unit 1110 performs the steps according to various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification. For example, the processing unit 1110 can perform the following steps: Figure 1 In step S110 shown in the figure, a multidimensional feature data set corresponding to the service request is obtained, wherein the multidimensional feature data set includes multiple feature data of multiple dimensions; in step S120, a feature vector set of the multidimensional feature data set is obtained, and the feature vector set of the multidimensional feature data set is input into a category prediction model to obtain a predicted category corresponding to the multidimensional feature data set, and a predicted probability corresponding to the predicted category; in step S130, target category information corresponding to the service request is obtained according to the predicted category and the predicted probability corresponding to the predicted category.

[0195] The storage unit 1120 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 11201 and / or a cache storage unit 11202 , and may further include a read-only storage unit (ROM) 11203 .

[0196] The storage unit 1120 may also include a program / utility 11204 having a set (at least one) of program modules 11205, such program modules 11205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0197] Bus 1130 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0198] The electronic device 1100 may also communicate with one or more external devices 1300 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable viewers to interact with the electronic device 1100, and / or communicate with any device that enables the electronic device 1100 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1150. Furthermore, the electronic device 1100 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 1160. As shown, the network adapter 1160 communicates with other modules of the electronic device 1100 via a bus 1130. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1100, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0199] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0200] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible implementations, various aspects of the present invention can also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of the present specification.

[0201] refer to Fig.12 As shown, a program product 1200 for implementing the above method according to an embodiment of the present invention is described, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0202] The program product may use 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 device, or any combination of the above. More specific examples (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 of the above.

[0203] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Readable signal media may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0204] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0205] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and 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 separate 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 through 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., through the Internet using an Internet service provider).

[0206] In addition, the above-mentioned figures are only schematic illustrations of the processes included in the method according to an exemplary embodiment of the present invention, and are not intended to be limiting. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be performed synchronously or asynchronously, for example, in multiple modules.

[0207] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

[0208] 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. A method for obtaining information, characterized in that: include: Acquire a multidimensional feature data set corresponding to a service request, wherein the multidimensional feature data set includes multiple feature data in multiple dimensions; the service request is triggered and generated by a customer service staff of a customer service center; Acquire a feature vector set of a multidimensional feature data set, and input the feature vector set of the multidimensional feature data set into a department category prediction model to obtain a predicted department category corresponding to the multidimensional feature data set, and a first prediction probability corresponding to the predicted department category, and input the feature vector set of the multidimensional feature data set and the predicted department category into a business category prediction model to obtain a predicted business category corresponding to the multidimensional feature data set, and a second prediction probability corresponding to the predicted business category; Obtain a target prediction business category according to the second prediction probability corresponding to each of the prediction business categories; determine whether the target prediction business category and the target prediction department category corresponding to the target prediction business category have an affiliation according to a preset affiliation; if so, configure the target prediction business category and the target prediction department category as the target category information corresponding to the service request; wherein the target prediction department category is determined according to the first prediction probability corresponding to the prediction department category.

2. The information acquisition method according to claim 1, characterized in that: The forecast sector categories include multiple; Inputting the feature vector set of the multidimensional feature data set and the predicted department category into the business category prediction model includes: Sorting the plurality of forecast department categories according to the first forecast probabilities corresponding to the forecast department categories, and acquiring a preset number of the forecast department categories as target forecast department categories; The data vector of the target prediction department category is obtained, the data vector of the target prediction department category is fused with the feature vector set of the multidimensional feature data set to obtain a fused feature vector set, and the fused feature vector set is input into the business category prediction model.

3. The information acquisition method according to claim 1, characterized in that: Get the feature vector set of a multidimensional feature dataset, including: The multidimensional feature data set is input into a vector extraction model to obtain a feature vector set corresponding to the multidimensional feature data set.

4. The information acquisition method according to claim 3, characterized in that: The feature data in the multidimensional feature data set includes discrete feature data, continuous feature data, multi-valued discrete feature data, and text feature data; The method further comprises: If the feature data is the discrete feature data, the continuous feature data, and the multi-valued discrete feature data, configuring the vector extraction model as a feature outline; If the feature data is the text feature data, the vector extraction model is configured as a word vector model.

5. The information acquisition method according to claim 4, characterized in that: Inputting the multidimensional feature data set into a vector extraction model to obtain a feature vector set corresponding to the multidimensional feature data set, including: Determining whether the feature data in the multidimensional feature data set is the text feature data; If yes, input the feature data into the word vector model to obtain a first feature vector corresponding to the feature data; If not, inputting the feature data into the feature outline to obtain a second feature vector corresponding to the feature data; The first feature vector and the second feature vector are concatenated to obtain a feature vector set corresponding to the multi-dimensional feature data set.

6. The information acquisition method according to claim 4, characterized in that: The word vector model includes three convolutional neural network layers, three maximum pooling layers, and one fully connected layer.

7. The information acquisition method according to claim 4, characterized in that: The method further comprises: Acquire all text data in the resource conversion platform, and configure the text data as text data samples; The word vector model to be trained is trained using the text data samples to obtain the word vector model.

8. The information acquisition method according to claim 1, characterized in that: The method further comprises: Acquire a historical multidimensional feature data set corresponding to a historical service request, and obtain a historical feature vector set corresponding to the historical multidimensional feature data set according to a vector extraction model; Configuring the historical feature vector set as a feature vector set sample, and configuring the target category information corresponding to the historical service request as a category label of the feature vector set sample; The category prediction model to be trained is trained by using the feature vector set samples and the category labels of the feature vector set samples to obtain the category prediction model.

9. The information acquisition method according to claim 8, characterized in that: The to-be-trained category prediction model is trained by using the feature vector set samples and the category labels of the feature vector set samples to obtain the category prediction model, including: Inputting the feature vector set samples into the to-be-trained category prediction model to obtain sample categories corresponding to the feature vector set samples; A loss function is determined according to the sample category and the category label, and the category prediction model is obtained by adjusting the parameters of the category prediction model to be trained until the loss function reaches a minimum.

10. The information acquisition method according to claim 9, characterized in that: The category prediction model to be trained includes a department category prediction model to be trained and a business category prediction model to be trained; Inputting the feature vector set samples into the to-be-trained category prediction model to obtain sample categories corresponding to the feature vector set samples includes: Inputting the feature vector set samples into the to-be-trained department category prediction model to obtain a plurality of sample department categories corresponding to the feature vector set samples, and a first probability corresponding to each of the sample department categories; Determine a target sample department category from the plurality of sample department categories according to each of the first probabilities; The feature vector set samples and the target sample department categories are input into the business category prediction model to be trained to obtain a plurality of sample business categories corresponding to the feature vector set samples and a second probability corresponding to each of the sample business categories.

11. The information acquisition method according to claim 10, characterized in that: The category labels include department category labels and business category labels; Determining a loss function according to the sample category and the category label, and adjusting the parameters of the to-be-trained category prediction model until the loss function reaches a minimum to obtain the category prediction model, including: Determine a first loss function according to the sample department category and the department category label, and obtain the department category prediction model by adjusting the parameters of the department category prediction model to be trained until the first loss function reaches a minimum; And determining a second rectification loss function according to the sample business category and the business category label, and obtaining the business category prediction model by adjusting the parameters of the business category prediction model to be trained until the second rectification loss function reaches a minimum.

12. The information acquisition method according to claim 1, characterized in that: The method further comprises: Sending the target category information corresponding to the service request to the service window of the terminal, and receiving a correction operation for the target category information, wherein the correction operation includes real category information; Determine a prediction accuracy rate according to the target category information and the real category information, and judge whether the prediction accuracy rate is less than a first accuracy rate threshold; If so, the multidimensional feature data set is used as a training sample, and the true category information is used as a label corresponding to the training sample. The category prediction model is trained through the training sample and the label corresponding to the training sample to obtain an updated category prediction model.

13. The information acquisition method according to claim 12, characterized in that: The method further comprises: configuring a flow control switch, and distributing the plurality of multidimensional feature data sets to the category prediction model and the updated category prediction model through the flow control switch; The prediction accuracy of the updated category prediction model is monitored in real time, and when the prediction accuracy of the updated category prediction model is greater than a second accuracy threshold, the category prediction model is replaced by the updated category prediction model.

14. The information acquisition method according to claim 1, characterized in that: Get the multi-dimensional feature data set corresponding to the service request, including: Real-time data interfaces of multiple dimensions are called, and multiple feature data corresponding to the service request are obtained through each of the real-time data interfaces.

15. An information acquisition device, characterized in that: include: A data acquisition module is used to acquire a multi-dimensional feature data set corresponding to a service request, wherein the multi-dimensional feature data set includes multiple feature data of multiple dimensions; the service request is triggered and generated by a customer service staff of a customer service center; A category prediction model is used to obtain a feature vector set of a multidimensional feature data set, and input the feature vector set of the multidimensional feature data set into a department category prediction model to obtain a predicted department category corresponding to the multidimensional feature data set, and a first prediction probability corresponding to the predicted department category, and input the feature vector set of the multidimensional feature data set and the predicted department category into a business category prediction model to obtain a predicted business category corresponding to the multidimensional feature data set, and a second prediction probability corresponding to the predicted business category; An information acquisition module is used to obtain a target prediction business category according to a second prediction probability corresponding to each of the prediction business categories; determine whether the target prediction business category and the target prediction department category corresponding to the target prediction business category have an affiliation according to a preset affiliation; if so, configure the target prediction business category and the target prediction department category as the target category information corresponding to the service request; wherein the target prediction department category is determined according to the first prediction probability corresponding to the prediction department category.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the information acquisition method according to any one of claims 1 to 14 is implemented.

17. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the information acquisition method as described in any one of claims 1 to 14.

Images