Method and device for determining override user

Abstract

The invention discloses a method and a device for determining override users, and relates to the technical field of computers. One specific embodiment of the method comprises the steps of obtaining non-real-time structured features and real-time interaction features of a user, performing feature coding on the two types of features to obtain a coding vector, performing calculation based on the coding vector to obtain a predicted override probability of the user, and further determining an override user based on the predicted override probability. According to the embodiment, the override probability of the user can be predicted in combination with the real-time data of the user, and the override user is finally determined, so that the accuracy of determining the override user is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method and device for determining a skip-level user. Background Art

[0002] At present, with the rapid development of Internet technology, e-commerce business has also been rapidly expanded, making online shopping an indispensable part of people's daily life. In the process of online shopping, consumers will inevitably encounter problems with product quality or service quality. When encountering problems, consumers will communicate with the merchant's customer service and try to solve them. If the merchant's customer service cannot solve them, consumers will escalate the problem to the platform's customer service for intervention and resolution. During the communication process with the platform's customer service, if the consumer is not satisfied with the solution given by the platform's customer service, he or she may choose to upgrade the problem. For possible upgrades, the e-commerce platform needs to take corresponding measures to provide early warning to reduce the number of upgrades. The specific method is generally that the e-commerce platform obtains customers who may skip levels through user data screening, formulates screening rules through manually selected features, and then implements customer screening.

[0003] However, the method of constructing screening rules by manually selecting features has certain limitations. It relies on manually pre-selected features and set thresholds. The features and thresholds are set based on historical business experience and are positively correlated with the business level of personnel. On the other hand, the ability to construct rules manually is limited. Rules can only be constructed based on numerical features, and features of other modal types cannot be effectively utilized. Summary of the invention

[0004] In view of this, the embodiments of the present invention provide a method and device for determining a skip-level user, which can predict the skip-level user in combination with real-time data, thereby improving the accuracy of determining the skip-level user.

[0005] To achieve the above object, according to one aspect of an embodiment of the present invention, a method for determining a skip-level user is provided, comprising:

[0006] Obtain the user's non-real-time structured features and real-time interactive features. The non-real-time structured features are constructed based on static structured information and historical interactive information, and the real-time interactive features are constructed based on real-time interactive information.

[0007] Perform feature encoding on non-real-time structured features and real-time interactive features to obtain the encoding vector corresponding to the user;

[0008] The predicted level skipping probability of the user is calculated based on the coding vector;

[0009] Determine the skip-level users based on the predicted probability of skipping the level.

[0010] Optionally, the static structured information includes user information and order information; before obtaining the user's non-real-time structured features, the method also includes: obtaining user information, order information and historical interaction information, and dividing them into discrete data and continuous data according to the attribute characteristics of the user information, order information and historical interaction information; performing feature extraction on the discrete data and continuous data respectively to obtain discrete features and continuous features; and splicing the discrete features and continuous features to obtain the user's non-real-time structured features.

[0011] Optionally, the real-time interaction information includes real-time text information, and the real-time interaction features include real-time text features. Before obtaining the user's real-time interaction features, the method also includes: segmenting the real-time text information to obtain multiple segmented segments; for each segmented segment, mapping and encoding each character in the segmented segment to obtain text features of the segmented segment; and splicing the text features of multiple segmented segments to obtain real-time text features.

[0012] Optionally, the real-time interaction information includes real-time voice information, and the real-time interaction features include real-time voice features. Before obtaining the user's real-time interaction features, the method also includes: dividing the real-time voice information into multiple voice segments according to a preset length; generating a frequency domain signal for each voice segment through Fourier transform; and performing feature extraction based on the frequency domain signal to obtain the real-time voice features.

[0013] Optionally, feature encoding is performed on non-real-time structured features and real-time interaction features to obtain a coding vector corresponding to the user, including: concatenating non-real-time structured features and real-time interaction features to obtain user embedded features; inputting the user embedded features into multiple coding layers for feature encoding, and summing the hidden states output by the multiple coding layers to obtain the coding vector corresponding to the user.

[0014] Optionally, before calculating the predicted level skipping probability of the user based on the coding vector, the method also includes: obtaining sample user data, the sample user data including sample user information, sample user order information and sample user interaction information, the sample user data including positive sample user data and negative sample user data; performing negative sampling on the negative sample user data by random sampling without replacement to obtain a sample user data set that meets a preset positive and negative sample ratio; performing feature construction based on the data of the sample user data set to obtain sample features; performing model training based on the sample features to generate a level skipping probability prediction model, and the model is used to generate the predicted level skipping probability of the user.

[0015] According to another aspect of an embodiment of the present invention, a device for determining a skip-level user is provided, including:

[0016] An acquisition module is used to acquire the user's non-real-time structured features and real-time interactive features. The non-real-time structured features are constructed based on static structured information and historical interactive information, and the real-time interactive features are constructed based on real-time interactive information.

[0017] The encoding module is used to perform feature encoding on the non-real-time structured features and the real-time interactive features to obtain the encoding vector corresponding to the user;

[0018] A calculation module, used for calculating the predicted level skipping probability of the user based on the coding vector;

[0019] The determination module is used to determine the skip-level user based on the skip-level prediction probability.

[0020] According to another aspect of an embodiment of the present invention, there is provided an electronic device, 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 method for determining a skip-level user provided in an embodiment of the present invention.

[0021] According to another aspect of an embodiment of the present invention, a computer readable medium is provided, on which a computer program is stored. When the program is executed by a processor, the method for determining a skip-level user provided in an embodiment of the present invention is implemented.

[0022] According to another aspect of the embodiments of the present invention, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the method for determining a skip-level user provided by the embodiments of the present invention is implemented.

[0023] One embodiment of the above invention has the following advantages or beneficial effects: it can construct features based on the user's non-real-time data in combination with real-time interaction data, encode the obtained features to obtain a coding vector, and predict the probability of user skipping based on the coding vector, thereby determining the skipping user. By combining real-time data and non-real-time data to perform skipping prediction, the timeliness and accuracy of determining the skipping user are improved, the response efficiency and service quality of customer service are significantly improved, and customer satisfaction and trust are further improved.

[0024] The further effects of the above-mentioned non-conventional optional manner will be described below in conjunction with the specific implementation manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The accompanying drawings are used to better understand the present invention and do not constitute an improper limitation of the present invention.

[0026] Figure 1 is a schematic diagram of main steps of a method for determining a skip-level user according to an embodiment of the present invention;

[0027] Figure 2 is a schematic diagram of the structure of the coding layer provided by an embodiment of the present invention;

[0028] Figure 3 It is an overall architecture diagram of the method for determining a skip-level user of the present invention;

[0029] Figure 4 2 is a schematic diagram of main modules of a device for determining a skip-level user according to an embodiment of the present invention;

[0030] Figure 5 is an exemplary system architecture diagram to which embodiments of the present invention may be applied;

[0031] Figure 6 It is a schematic diagram of the structure of a computer system of a terminal device or a server suitable for implementing an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following is a description of exemplary embodiments of the present invention in conjunction with the accompanying drawings, including various details of the embodiments of the present invention to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present invention. Similarly, for clarity and conciseness, the description of well-known functions and structures is omitted in the following description.

[0033] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution disclosed in the present invention are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0034] It should be noted that in the technical solution of the present invention, the collection, use, storage, sharing and transfer of user personal information involved are in compliance with the provisions of relevant laws and regulations, and it is necessary to inform the user and obtain the user's consent or authorization. When applicable, the user's personal information is de-identified and / or anonymized and / or encrypted.

[0035] Figure 1 FIG. 4 is a schematic diagram of the main steps of a method for determining a skip-level user according to an embodiment of the present invention. Figure 1 As shown, the method for determining a skip-level user in the embodiment of the present invention includes steps S101 to S104.

[0036] Step S101, obtaining the user's non-real-time structured features and real-time interactive features, wherein the non-real-time structured features are constructed based on static structured information and historical interactive information, and the real-time interactive features are constructed based on real-time interactive information.

[0037] Obtain non-real-time structured features based on user data and real-time interactive features based on user real-time interactive data. The non-real-time structured features are constructed based on the user's static structured information (e.g., user type, user order amount, user order type, etc.) and historical interactive information (e.g., historical communication times with customer service, communication methods, communication results, etc.); real-time interactive features are constructed based on real-time interactive information (including real-time voice, real-time text, etc.).

[0038] Optionally, the selection of specific static structured information, historical interaction information and real-time interaction information can be determined according to actual conditions, and there is no limitation to this.

[0039] Step S102, feature encoding is performed on the non-real-time structured features and the real-time interactive features to obtain a coding vector corresponding to the user.

[0040] After obtaining the constructed features, the features need to be encoded. First, the constructed features are concatenated, and then the concatenated features are encoded. The encoding operation can be implemented through a certain number of encoders (encoding layers). After the features are input into the encoding layer, the hidden state output by each encoder is taken out and summed to obtain the encoding vector output by the final encoding layer. Among them, the number of encoding layers is generally 12, which has the best effect.

[0041] Step S103, calculating the predicted level skipping probability of the user based on the coding vector.

[0042] The predicted probability of the user's leapfrogging is calculated and predicted based on the obtained encoding vector. The specific process can be implemented through a pre-trained prediction model. The prediction model includes a bidirectional LSTM (Long Short-Term Memory Network), a fully connected layer, a SoftMax (Normalized Exponential Function) layer, and a probability correction unit. After the encoding vector is input into the prediction model, the initial probability is obtained through prediction, and the probability correction unit is further used to correct the probability to obtain the user's predicted leapfrogging probability.

[0043] Step S104: determining a skip-level user based on the skip-level prediction probability.

[0044] The skipping user is determined according to the obtained skipping prediction probability. Specifically, the obtained prediction value is compared with a preset skipping probability threshold, and the user whose probability is greater than the skipping probability threshold is determined as the skipping user. The skipping user is a user who is undergoing an upgrade process.

[0045] According to the method for determining a skip-level user provided by an embodiment of the present invention, it is possible to construct features based on the user's non-real-time data in combination with real-time interaction data, encode the obtained features to obtain a coding vector, and predict the probability of the user's skipping level based on the coding vector, thereby determining the skip-level user. By combining real-time data and non-real-time data for skip-level prediction, the timeliness and accuracy of determining the skip-level user are improved, the response efficiency and service quality of customer service are significantly improved, and customer satisfaction and trust are further improved.

[0046] Optionally, the static structured information includes user information and order information; before obtaining the user's non-real-time structured features, the method also includes: obtaining user information, order information and historical interaction information, and dividing them into discrete data and continuous data according to the attribute characteristics of the user information, order information and historical interaction information; performing feature extraction on the discrete data and continuous data respectively to obtain discrete features and continuous features; and splicing the discrete features and continuous features to obtain the user's non-real-time structured features.

[0047] The static structured information obtained mainly includes user information and order information. User information includes the province and city of the user's delivery address, member value, user type, etc.; order information includes the amount after the order discount, product sku (unique identifier of the product), the first-level category, second-level category, third-level category of the product, order type, after-sales service order link, compensation review status, etc. Among them, different user classifications include: upgrade processing risk users, indicating that users have been upgraded in the past period of time; users with too many returns and exchanges recently; users with good reputation, etc. Historical interaction information includes historical interaction data records between customers and customer service, including the number of incoming calls, the number of instant messaging contacts, the number of unreachable contacts, the number of negative reviews, the number of reprocessing, the number of reports and hang-ups, etc. Based on the above-obtained user information, order information and historical interaction information, the information type is judged, that is, whether it is discrete data or continuous data. According to the information type, the above information is feature extracted to obtain discrete features and continuous features. Each type of feature obtained is spliced ​​separately, and finally the discrete features and continuous features are spliced ​​to obtain the user's non-real-time structured features.

[0048] Among them, the characteristics of each information are as follows. User information characteristics: delivery address province and city, membership value, user type, etc. Order information characteristics: order amount after discount, product sku, first-level category, second-level category, third-level category, order type, after-sales service order link, compensation review status, price insurance application status, invoice application status, service order processing results. Historical interaction information characteristics: number of incoming calls, number of contacts, number of unanswered calls, number of negative reviews, number of reports and hang-ups, number of repeated form submissions, number of deterioration in communication situations, number of work order initiations, number of round trips, number of transfers, number of order messages, length of time the event has not been followed up, whether an agreement has been reached, and whether the promise has not been fulfilled.

[0049] Optionally, the two types of features are constructed and concatenated as follows. Discrete features are encoded from 1 to N according to frequency, and continuous features are standardized using Z-Score. Discrete features and continuous features are then encoded differently. Discrete features: 1) Discrete features with a small number of categories (<20) are first one-hot encoded and then mapped to a linear layer. =50 dimensions; 2) Discrete features with a large number of categories (>=20) are directly encoded through an Embedding layer as = 50 dimensions. There are 20 categories of discrete features in total. These discrete feature codes are concatenated to obtain the final discrete features with a total of 1000 dimensions. Continuous features: Continuous features are first bucketed. For example, for the number of incoming calls, 0 to 1 is encoded as 1, 2 to 3 is encoded as 2, 3 to 4 is encoded as 3, and so on. Then the continuous features after digital encoding are encoded through an Embedding layer as = 50 dimensions. There are 15 types of continuous features in total. These continuous feature codes are concatenated to obtain the final continuous features with a total dimension of 750. Finally, the discrete features and continuous features are concatenated to obtain a 1750-dimensional structured feature vector, which is then mapped to 768 dimensions through a linear layer. The aligned dimensions facilitate concatenation with subsequent feature vectors.

[0050] Optionally, the real-time interaction information includes real-time text information, and the real-time interaction features include real-time text features. Before obtaining the user's real-time interaction features, the method also includes: segmenting the real-time text information to obtain multiple segmented segments; for each segmented segment, mapping and encoding each character in the segmented segment to obtain text features of the segmented segment; and splicing the text features of multiple segmented segments to obtain real-time text features.

[0051] The user's real-time interaction information includes real-time text information, and the corresponding features include real-time text features. Specifically, the real-time text information is all the text information obtained at the end of the chat between the customer and the customer service. The construction process of the real-time text feature is as follows. First, use the WordPiece (word segmentation) segmenter of the Bert (a language representation model) pre-trained model to segment the text to obtain multiple segmented segments. Then use Bert's vocabulary vocab to map each word to a digital code. For example, "I love my motherland" is mapped to "[2769, 4263, 704, 1744]" in the Bert vocabulary to obtain the text features of the slice segment. The features of each segment are spliced ​​to obtain the real-time text feature. Finally, it can be encoded through the Embedding layer as = 768 dimensions and add it to the positional and fragment encodings.

[0052] Optionally, the real-time interaction information includes real-time voice information, and the real-time interaction features include real-time voice features. Before obtaining the user's real-time interaction features, the method also includes: dividing the real-time voice information into multiple voice segments according to a preset length; generating a frequency domain signal for each voice segment through Fourier transform; and performing feature extraction based on the frequency domain signal to obtain the real-time voice features.

[0053] In addition to real-time text information, real-time interactive information can also include real-time voice information, that is, real-time interactive information can include only real-time text information or real-time voice information, or both real-time text information and real-time voice information. When real-time voice information is included, it is necessary to construct corresponding real-time voice features. The voice features are constructed as follows. First, obtain the voice signal of the recorded call between the user and the customer service, and divide the signal into frames according to the length of each frame of 15ms to obtain multiple voice segments. Apply a window function (Hamming window, etc.) to each voice segment, and after Fourier transform, convert the voice signal into a frequency domain signal, and perform feature extraction on it to obtain real-time voice features. Among them, the feature extraction operation includes: passing the frequency domain signal through a set of Mel filters to obtain the output energy of each filter; taking the logarithm of the output energy of each filter; performing discrete cosine transform on the logarithmic energy value of each filter to extract the frequency features and obtain the MFCC (Mel filter) coefficients. Finally, the extracted features can also be normalized and input into the GRU (recurrent neural network) to be encoded as =768 dimensions.

[0054] Optionally, the real-time interaction information may also include user operation behavior information, such as the user's behavior of continuing to browse other products, exiting the current page, etc.

[0055] Optionally, feature encoding is performed on non-real-time structured features and real-time interaction features to obtain a coding vector corresponding to the user, including: concatenating non-real-time structured features and real-time interaction features to obtain user embedding features; inputting user embedding features into multiple coding layers for feature encoding, and summing up the hidden states output by multiple coding layers to obtain a coding vector corresponding to the user. When performing feature encoding, it is first necessary to concatenate non-real-time structured features and real-time interaction features, and obtain user embedding features through a fully connected layer. Specifically, the dimensions of the three embeddings, speech embedding, text embedding, and structured data embedding, are all 768 dimensions. After concatenation, they are input into the fully connected layer for dimensional transformation to obtain input embedding, with an input dimension of 768*3 and an output dimension of 768. The user embedding features are transformed through 12 steps such as Figure 2 The encoding layers shown (including Add&norm (residual connection and normalization), Feed forward (feedforward neural network), Multi-head attention (multi-head attention)) perform feature encoding, and the hidden states output by multiple encoding layers are summed to obtain the encoding vector corresponding to the user.

[0056] Optionally, before obtaining the predicted level-skipping probability of the user based on the coding vector calculation, the method further includes: obtaining sample user data, the sample user data including sample user information, sample user order information and sample user interaction information, and the sample user data including positive sample user data and negative sample user data; performing negative sampling on the negative sample user data by random sampling without replacement to obtain a sample user data set that meets the preset positive and negative sample ratio; constructing features based on the data of the sample user data set to obtain sample features; training a model based on the sample features to generate a level-skipping probability prediction model, which is used to generate the predicted level-skipping probability of the user. Before performing level-skipping probability prediction, it is necessary to train the level-skipping probability prediction model. First, obtain sample user data, which includes three types of data that need to be trained: user information, user order information and sample user interaction information. The sample data includes positive sample data and negative sample data, that is, data of level-skipping users and data of non-level-skipping users. Since the amount of real skip-level data accounts for a very low proportion of the total after-sales service orders, only about 1%, negative samples are negatively sampled. The present invention uses the method of randomly extracting negative samples without replacement in the negative sample data set so that the final positive and negative sample ratio is 1:3, and obtains a sample user data set. Based on the sample user data set, training is performed first, and features are constructed to obtain sample features. Based on the sample features, training is performed to obtain a skip-level probability prediction model.

[0057] Optionally, the probability correction formula of the leapfrogging probability prediction model is:

[0058]

[0059] Among them, ω represents the sampling ratio of negative samples (for example, 3 / 100), is the initial probability of the output, is the corrected probability. If the value is ≥0.3, it is predicted to be a skip-level user, and the customer service staff will be reminded that the user may be at risk of being upgraded. <0.3, the prediction is normal. The sampling ratio of negative samples represented by ω can be set according to the actual situation and there is no restriction on this.

[0060] Optionally, after the skipped user is identified, the user may be processed according to a preset processing method, for example, timely intervention by professionals, etc., to improve the user experience and reduce the probability of being upgraded.

[0061] The method for determining the skipping user provided by the embodiment of the present invention can combine the two types of data, real-time data and non-real-time data, to predict the probability of the user skipping. Specifically, it includes: on the basis of the user's non-real-time data, combining the real-time interaction data to construct features, and encoding the obtained features to obtain a coding vector, and predicting the probability of the user skipping based on the coding vector, and then determining the skipping user. By combining real-time data and non-real-time data to make a skipping prediction, the accuracy and timeliness of the determination of the skipping user are significantly improved, which helps to handle the skipping user in a timely manner, improve user satisfaction, and reduce the probability of upgrade processing.

[0062] like Figure 3 The figure shows the overall architecture of the method for determining the skip-level user of the present invention. It includes preprocessing the voice signal and passing it through the GRU network to obtain voice features (voice embedding); extracting features from the user's real-time text to obtain text features (text embedding); extracting features from user information, order information and historical interaction information to obtain structured data features (structured data embedding). After splicing the three types of features, pass them through the fully connected layer, input the output feature vector into the encoding layer, and obtain the encoding vector by taking the hidden state vector and summing them. After inputting the encoding vector into the skip-level probability prediction model, the output prediction probability is obtained.

[0063] Figure 4 4 is a schematic diagram of a device 400 for determining a skip-level user provided by an embodiment of the present invention. Figure 4 As shown, the device 400 for determining a skip-level user includes an acquisition module 401 , an encoding module 402 , a calculation module 403 and a determination module 404 .

[0064] The acquisition module 401 is used to acquire the non-real-time structured features and real-time interactive features of the user, wherein the non-real-time structured features are constructed based on static structured information and historical interactive information, and the real-time interactive features are constructed based on real-time interactive information;

[0065] The encoding module 402 is used to perform feature encoding on the non-real-time structured features and the real-time interactive features to obtain an encoding vector corresponding to the user;

[0066] A calculation module 403, configured to calculate a predicted level skipping probability of a user based on the coding vector;

[0067] The determination module 404 is used to determine the skip-level user based on the skip-level prediction probability.

[0068] According to the device for determining the skipped user provided by the embodiment of the present invention, it is possible to construct features based on the user's non-real-time data in combination with real-time interaction data, encode the obtained features to obtain a coding vector, and predict the probability of the user's skipping level based on the coding vector, thereby determining the skipped user. By combining real-time data and non-real-time data for skipping level prediction, the timeliness and accuracy of the determination of the skipped user are improved, the response efficiency and service quality of customer service are significantly improved, and the customer satisfaction and trust are further improved.

[0069] Optionally, the static structured information includes user information and order information; the device also includes: an acquisition module 401, used to acquire user information, order information and historical interaction information, and divide the user information, order information and historical interaction information into discrete data and continuous data according to the attribute characteristics of the user information, order information and historical interaction information; a feature extraction module 405 (not shown in the figure), used to extract features from discrete data and continuous data respectively to obtain discrete features and continuous features; a splicing module 406, used to splice discrete features and continuous features to obtain non-real-time structured features of the user.

[0070] Optionally, the device 400 also includes: a segmentation module 407 (not shown in the figure), which is used to segment the real-time text information to obtain multiple segmented segments; an encoding module 402, which is used to map and encode each character in each segmented segment to obtain text features of the segmented segment; and a splicing module 406, which is used to splice the text features of multiple segmented segments to obtain real-time text features.

[0071] Optionally, the device 400 also includes: a division module 408 (not shown in the figure), which is used to divide the real-time voice information into multiple voice segments according to a preset length; a transformation module 409 (not shown in the figure), which is used to generate a frequency domain signal through Fourier transform for each voice segment; and a feature extraction module 405, which is used to perform feature extraction based on the frequency domain signal to obtain real-time voice features.

[0072] Optionally, the encoding module 402 can be specifically used to: concatenate non-real-time structured features and real-time interaction features to obtain user embedded features; input the user embedded features into multiple encoding layers for feature encoding, and sum the hidden states output by the multiple encoding layers to obtain the encoding vector corresponding to the user.

[0073] Optionally, the device 400 also includes: an acquisition module 401, used to acquire sample user data, the sample user data includes sample user information, sample user order information and sample user interaction information, and the sample user data includes positive sample user data and negative sample user data; a sampling module 410 (not shown in the figure), used to perform negative sampling by randomly selecting the negative sample user data without replacement, so as to obtain a sample user data set that satisfies a preset positive and negative sample ratio; a construction module 411 (not shown in the figure), used to perform feature construction based on the data of the sample user data set to obtain sample features; a training module 412 (not shown in the figure), used to perform model training based on the sample features to generate a leapfrogging probability prediction model, and the model is used to generate a predicted leapfrogging probability for the user.

[0074] The device for determining the skipping user provided in the embodiment of the present invention can combine the two types of data, real-time data and non-real-time data, to predict the probability of the user skipping the level. Specifically, it includes: on the basis of the user's non-real-time data, combining the real-time interaction data to construct features, and encoding the obtained features to obtain a coding vector, and predicting the probability of the user skipping the level based on the coding vector, and then determining the skipping user. By combining real-time data and non-real-time data to make a skipping prediction, the accuracy and timeliness of the determination of the skipping user is significantly improved, which helps to handle the skipping user in a timely manner, improve user satisfaction, and reduce the probability of upgrade processing.

[0075] Figure 5 An exemplary system architecture 500 is shown to which the method for determining a skip-level user or the apparatus for determining a skip-level user according to the embodiments of the present invention can be applied.

[0076] like Figure 5 As shown, system architecture 500 may include terminal devices 501, 502, 503, a network 504 and a server 505. Network 504 is used to provide a medium for communication links between terminal devices 501, 502, 503 and server 505. Network 504 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc.

[0077] Users can use terminal devices 501, 502, 503 to interact with server 505 through network 504 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 501, 502, 503, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only for example).

[0078] The terminal devices 501 , 502 , and 503 may be various electronic devices having a display screen and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0079] The server 505 may be a server that provides various services, such as a backend management server (only an example) that provides support for shopping websites browsed by users using the terminal devices 501, 502, and 503. The backend management server may analyze and process the received prediction request and other data, and feed back the processing results (such as skipped users - only an example) to the terminal device.

[0080] It should be noted that the method for determining the skip-level user provided in the embodiment of the present invention is generally executed by the server 505 , and accordingly, the device for determining the skip-level user is generally arranged in the server 505 .

[0081] It should be understood that Figure 5 The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0082] Reference below Figure 6 , which shows a schematic diagram of the structure of a computer system 600 of a terminal device or a server suitable for implementing an embodiment of the present invention. Figure 6 The terminal device or server 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.

[0083] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the system 600 are also stored. The CPU 601, the ROM 602, and the RAM 603 are connected to each other through a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0084] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage section 608 as needed.

[0085] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the central processing unit (CPU) 601, the above-mentioned functions defined in the system of the present invention are executed.

[0086] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-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 of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer 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. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0087] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0088] The units or modules involved in the embodiments of the present invention may be implemented in software or hardware. The units or modules described may also be arranged in a processor, for example, they may be described as: a processor includes an acquisition module, an encoding module, a calculation module, and a determination module. The names of these units or modules do not, in some cases, constitute limitations on the units or modules themselves, for example, the determination module may also be described as a "module for determining a skip-level user based on the predicted probability of skip-level".

[0089] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes:

[0090] Obtain the user's non-real-time structured features and real-time interactive features. The non-real-time structured features are constructed based on static structured information and historical interactive information, and the real-time interactive features are constructed based on real-time interactive information.

[0091] Perform feature encoding on non-real-time structured features and real-time interactive features to obtain the encoding vector corresponding to the user;

[0092] The predicted level skipping probability of the user is calculated based on the coding vector;

[0093] Determine the skip-level users based on the predicted probability of skipping the level.

[0094] According to the technical solution of the embodiment of the present invention, it is possible to combine two types of data, real-time data and non-real-time data, to predict the probability of user skipping. Specifically, it includes: based on the user's non-real-time data, combining real-time interaction data to construct features, and encoding the obtained features to obtain a coding vector, and predicting the probability of user skipping based on the coding vector, and then determining the skipping user. By combining real-time data and non-real-time data to make a skipping prediction, the accuracy and timeliness of the determination of the skipping user is significantly improved, which helps to handle the skipping user in a timely manner, improve user satisfaction, and reduce the probability of upgrade processing.

[0095] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for determining a skip-level user, characterized in that: include: Acquire non-real-time structured features and real-time interactive features of the user, wherein the non-real-time structured features are constructed based on static structured information and historical interactive information, and the real-time interactive features are constructed based on real-time interactive information; Performing feature encoding on the non-real-time structured feature and the real-time interactive feature to obtain a coding vector corresponding to the user; Calculate the predicted level skipping probability of the user based on the coding vector; A skip-level user is determined based on the skip-level prediction probability.

2. The method according to claim 1, characterized in that The static structured information includes user information and order information; Before obtaining the non-real-time structured features of the user, the method further includes: Acquire user information, order information, and historical interaction information, and divide the information into discrete data and continuous data according to the attribute characteristics of the user information, the order information, and the historical interaction information; Extracting features from the discrete data and the continuous data respectively to obtain discrete features and continuous features; The discrete features and the continuous features are concatenated to obtain non-real-time structured features of the user.

3. The method according to claim 1, characterized in that The real-time interaction information includes real-time text information, the real-time interaction feature includes a real-time text feature, and before acquiring the real-time interaction feature of the user, the method further includes: Segmenting the real-time text information to obtain a plurality of segmented segments; For each segmented segment, mapping and encoding each character in the segmented segment to obtain text features of the segmented segment; The text features of the multiple segmented segments are concatenated to obtain real-time text features.

4. The method according to any one of claims 1 to 3, characterized in that The real-time interaction information includes real-time voice information, the real-time interaction feature includes real-time voice feature, and before acquiring the real-time interaction feature of the user, the method further includes: Dividing the real-time voice information into a plurality of voice segments according to a preset length; Generate a frequency domain signal by Fourier transforming each speech segment; Feature extraction is performed based on the frequency domain signal to obtain real-time speech features.

5. The method according to claim 1, characterized in that The performing feature encoding on the non-real-time structured feature and the real-time interactive feature to obtain the encoding vector corresponding to the user includes: The non-real-time structured feature and the real-time interactive feature are concatenated to obtain a user embedded feature; The user embedding features are input into multiple encoding layers for feature encoding, and the hidden states output by the multiple encoding layers are summed to obtain the encoding vector corresponding to the user.

6. The method according to claim 1, characterized in that Before calculating the predicted level skipping probability of the user based on the coding vector, the method further includes: Acquire sample user data, the sample user data including sample user information, sample user order information and sample user interaction information, the sample user data including positive sample user data and negative sample user data; Negative sampling is performed by randomly selecting the negative sample user data without replacement to obtain a sample user data set that satisfies a preset positive and negative sample ratio; Constructing features based on the data of the sample user data set to obtain sample features; Model training is performed based on the sample features to generate a level skipping probability prediction model, where the model is used to generate a predicted level skipping probability for a user.

7. A device for determining a skip-level user, characterized in that: include: An acquisition module, used to acquire a user's non-real-time structured feature and a real-time interactive feature, wherein the non-real-time structured feature is constructed based on static structured information and historical interactive information, and the real-time interactive feature is constructed based on real-time interactive information; An encoding module, used for encoding the non-real-time structured features and the real-time interactive features to obtain an encoding vector corresponding to the user; A calculation module, used for calculating the predicted level skipping probability of the user based on the coding vector; A determination module is used to determine a skip-level user based on the skip-level prediction probability.

8. 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, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer readable medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images